[{ "id": "/notes/pubs/10-years/", "title": "10 Years of CV Debugging Lessons", "url": "/notes/pubs/10-years/", "tags": ["computer-vision","debugging","lessons-learned","opencv"], "hashtags": "#cv #debugging #lessonslearned #opencv", "category": "hub", "body": "\n 10 Years of CV Debugging Lessons — Lessons learned from a decade of debugging computer vision systems in production. — https://www.pirahansiah.com/notes/pubs/10-years/\n\n\nI engage directly with investors, customers, and companies to present our solutions and understand their concerns and product needs. By organizing and analyzing these requirements from a professional, technical perspective, I design and architect robust software systems. I translate these user needs into actionable user stories, project milestones, and tasks to deliver reliable solutions, while planning future optimizations and feature roadmaps.\nupdate Aug. 2026\n\nfor color detection HSV can use multiple colors so better and more robust\nusing multi color HSV during ROI selection for detect the multi objects\n\nMetis PCIe 4-AIPU card axelera \ndifferent position of camera \nshadow rejection \ni tested many models for Hermes for 4060 with 8gb gpu ram only need 2gb model + 6gb contecsx 64k so \nHermes: need 5gb for context , so the qwen3:4b is only 3gb and can run with it on 4060 laptop 8gb ram, so connect browser is working fine and fast so the ram is really important\n\nthe speed of 4060 vs 5060 gap gddr 6 vs gddr7\nthe apple use lines to increase bus ram for lpddr so achieve same with low ram\n\n#ComputerVision_Edge_AI\nHigh-Accuracy Real-Time Medical Image Processing on Embedded Systems\n#RealTimeEmbeddedMedicalImageProcessing\nC++:\ncv::setNumThreads(cv::getNumberOfCPUs(\nPython:\nimport multiprocessing\nnum_cpus = multiprocessing.cpu_count(\ncv2.setNumThreads(num_cpus\n\nAs an AI and computer vision expert with over a d..." },{ "id": "/notes/pubs/papers/2d-3d-mapping/", "title": "2D vs 3D Mapping for Moving Objects", "url": "/notes/pubs/papers/2d-3d-mapping/", "tags": [], "hashtags": "", "category": "hub", "body": "\n 2D vs 3D Mapping for Moving Objects — Comparison of 2D and 3D mapping approaches for environment movement detection. — https://www.pirahansiah.com/notes/pubs/papers/2d-3d-mapping/\n\n\ntags: 3d-mapping, object-detection, computer-vision, robotics]\nhashtags: “#3dmapping #objectdetection #cv #robotics”\n\nMy_Conference_Paper_2D_versus_3D_Map_for_Environment_Movement_Objects\n\nhttps://www.pirahansiah.com/notes/pubs/papers/conference-paper_2D_versus_3D_Map_for_Environment_Movement_Objects\n\nspotify\n\nPDF Download My Conference Paper\n\n\n\nPaper: 2D versus 3D Map for Environment Movement Objects\n\n1. Introduction\n\n Main Focus: The use of 2D vs. 3D mapping in tracking environmental movement objects.\n Key Applications:\n \n Autonomous systems\n Robotics\n \n \n Challenges:\n \n Object rotation issues in 2D mapping\n Scaling problems in 2D environments\n \n \n\n\n2. Materials and Methods\n\n 3D Mapping Algorithms:\n \n Introduces algorithms that manage:\n \n Scale\n Position\n Orientation\n \n \n \n \n Data Capture:\n \n Details the camera and sensor systems used to capture movement data.\n \n \n\n\n3. Experiments and Discussions\n\n System Performance Comparison:\n \n Evaluates 2D vs. 3D mapping systems based on:\n \n Accuracy\n Adaptability to different conditions\n \n \n Results indicate significant improvement in accuracy with the 3D system.\n \n \n Test Conditions:\n \n Different test scenarios were designed to assess:\n \n Real-world movement\n ..." },{ "id": "/notes/docs/cv/3d/", "title": "3D Vision & Real-Time Multi-Camera Systems", "url": "/notes/docs/cv/3d/", "tags": ["computer-vision","3d-vision","depth-sensing","multi-camera"], "hashtags": "#cv #3dvision #depthsensing #multicamera", "category": "hub", "body": "\n 3D Vision & Real-Time Multi-Camera Systems — Real-time 3D point cloud generation, depth sensing, and multi-camera synchronization for computer vision applications. — https://www.pirahansiah.com/notes/docs/cv/3d/\n\n\n\n MASt3R-SLAM: Real-Time Dense SLAM with 3D Reconstruction Priors\n \n\n\nReal-Time 3D Point Cloud Generation and Visualization from Depth Data\n\nThe fusion of depth sensing and 3D visualization opens remarkable possibilities for interactive applications. By converting 2D depth maps into 3D point clouds, we can build systems that bridge physical and digital realms in real-time.\n\nDepth to 3D Conversion\n\nThe foundation of this approach lies in the deprojection process - transforming pixel coordinates and their associated depth values into 3D space. This requires camera intrinsic parameters (focal length, principal point to perform the perspective transformation:\n\ndef deproject_point(u, v, depth, camera_matrix:\n fx = camera_matrix0, 0] # Focal length X\n fy = camera_matrix1, 1] # Focal length Y\n cx = camera_matrix0, 2] # Principal point X\n cy = camera_matrix1, 2] # Principal point Y\n \n # Convert to 3D coordinates\n x = (u - cx * depth / fx\n y = (v - cy * depth / fy\n z = depth\n \n return np.array(x, y, z]\n\n\nReal-Time Visualization Strategies\n\nVisualizing 3D data interactively requires threading to prevent blocking the main application loop. A separate thread can handle display updates while maintaining responsive input handling:\n\ndef start_visualizer_thread(:\n global visualizer_thread, visualizer_active\n visualizer_active..." },{ "id": "/404.html", "title": "Page Not Found", "url": "/404.html", "tags": [], "hashtags": "", "category": "hub", "body": "\n 404\n This page doesn't exist or has been moved.\n ← Back to Home\n\n" },{ "id": "/notes/pkm/AboutMe/", "title": "Dr. Farshid Pirahansiah — AI & Computer Vision Engineer", "url": "/notes/pkm/AboutMe/", "tags": [], "hashtags": "", "category": "pkm", "body": "\n\n\n\n\n\n\n \n // About Me\n Dr. FarshidPirahansiah\n \n Senior AI Engineer · Computer Vision ResearcherEdge AI Specialist · PhD · Author & Speaker\n \n \n 11+Years Experience\n 21Publications\n 3Patents\n 38%Inference Cost Cut\n \n \n \n // Professional Identity\n \n I build intelligent systems that see, understand, and act — from real-time edge AI on embedded hardware to large-scale MLOps pipelines in the cloud.\n \n With a PhD in Computer Vision and 12+ years bridging research and production, I specialise in taking AI from prototype to deployment — whether that's running embedded deep learning models on a Nvidia, Raspberry Pi, Axelera, Hailo AI chip or scaling medical imaging analysis to tens of thousands of patients daily.\n \n \n Computer Vision\n Edge AI\n Medical Imaging\n MLOps\n Embedded Systems\n 3D Reconstruction\n IoT\n Cloud ML\n \n \n\n\n\n\n \n \n 01\n My Story\n \"Innovation thrives where precision meets possibility.\"\n \n \n My journey into AI began with a fascination for how machines could be taught to perceive the physical world. During my PhD, I developed fuzzy logic systems for camera calibration and image segmentation — work that led to patents and set the foundation for a career at the intersection of theory and hardware.\n Over the years I have worked across academia, enterprise, and deep-tech — from MIMOS (Malaysia's national ICT research centre, where I built patented Intelligent Video Analytics systems, to medical device companies improving d..." },{ "id": "/notes/pkm/TOC/", "title": "PKM Methods & Website Structure", "url": "/notes/pkm/TOC/", "tags": ["pkm","knowledge-management","productivity"], "hashtags": "#pkm #knowledgemanagement #productivity", "category": "pkm", "body": "\n PKM Methods & Website Structure — Personal knowledge management approaches from top practitioners and site architecture — https://www.pirahansiah.com/notes/pkm/TOC/\n\n\nPARA, Zettelkasten, Second Brain, Link Your Thinking, LYT, Maps of Content, MOC, Johnny.Decimal, Atomic Notes, GTD, Getting Things Done, CODE, Pillars Pipelines Vaults, Inbox Method, Action Method, Luhmann System, Slipbox, Commonplace Book, Laboratory Method, Evergreen Notes, Digital Garden, The Archive, Bullet Journal, Accessory System, Folderless System, Atomic Thinking, Knowledge Graph, Semantic Linkage, Personal Information Management, PIM, Personal Research Management, PRM, Personal Knowledge Infrastructure, PKI, Information Scaffolding, Transactive Memory, Knowledge Crystallization, Cognitive Offloading, Distributed Cognition, Cybernetic PKM, Rhizomatic Learning, Memex, Project Xanadu, Sensemaking, Information Foraging Theory, Personal Digital Archiving, Curatorial Learning, Epistemic Foraging, Holigarchy, BASB, ACCESS, ACE, The Pillar Method, PAIN, REAP, QEC, SQ3R, PQ4R, Cornell Method, Feynman Technique, DiSSS, SQRRR, Morphological Analysis, Knowledge Graphing, Triple-Entry Accounting of Thought, Heuristic Systematic Model, PASK, Conversation Theory, Cynefin Framework, Double-Loop Learning, Metacognitive Scaffolding, Personal Learning Networks, PLN, Knowledge Work, Cognitive Mapping, Concept Mapping, Spatial Hypertext, Non-Linear Documentation, Multi-Dimensional Tagging, Hierarchical Foldering, Flat File System, Associative Trails, Navigational Orienting, Information Curation, Semantic MediaW..." },{ "id": "/notes/pubs/papers/adaptive-segmentation/", "title": "PSNR-Based Segmentation for LPR", "url": "/notes/pubs/papers/adaptive-segmentation/", "tags": ["image-segmentation","psnr","thresholding","license-plate-recognition"], "hashtags": "#imagesegmentation #psnr #thresholding #lpr", "category": "hub", "body": "\n PSNR-Based Segmentation for LPR — Adaptive image segmentation using PSNR for license plate recognition systems. — https://www.pirahansiah.com/notes/pubs/papers/adaptive-segmentation/\n\n\nAdaptive_Image_Segmentation_Based_on_PSNR_for_License_Plate_Recognition\n\nhttps://www.pirahansiah.com/notes/pubs/papers/Adaptive_Image_Segmentation_Based_on_PSNR_for_License_Plate_Recognition\n\nspotify\n\nPDF Download My Conference Paper\n\n\n\nAdaptive Image Segmentation Based on PSNR for License Plate Recognition\n\n1. Introduction\n\n Objective: Propose an adaptive threshold method using Peak Signal-to-Noise Ratio (PSNR.\n Applications of Image Segmentation:\n \n License Plate Recognition (LPR\n Preprocessing in image analysis\n Object detection and classification\n \n \n\n\n2. Pattern Recognition\n\n Applications:\n \n Optical Character Recognition (OCR\n Biometrics (e.g., face, fingerprint recognition\n Medical diagnostics (e.g., X-ray analysis\n Military applications (e.g., Automated Target Recognition, Image Segmentation\n \n \n\n\n3. Motivation\n\n Thresholding:\n \n Single and multi-level thresholding methods\n PSNR-based threshold for improved accuracy in segmentation\n \n \n Challenges:\n \n Differing environmental conditions (e.g., contrast, lighting\n Object complexity (e.g., fonts, colors, sizes\n \n \n\n\n4. Proposed Method\n\n PSNR as a Metric:\n \n Used to determine optimal threshold value\n Evaluated across various test cases (dark, medium, bright images\n \n \n Algorithm:\n \n Adaptive thresholding method steps:\n ..." },{ "id": "/notes/pubs/journals/adaptive-thresholding/", "title": "Adaptive Thresholding Based on PSNR", "url": "/notes/pubs/journals/adaptive-thresholding/", "tags": ["thresholding","psnr","adaptive-thresholding","image-segmentation"], "hashtags": "#thresholding #psnr #adaptivethresholding #imagesegmentation", "category": "hub", "body": "\n Adaptive Thresholding Based on PSNR — PSNR-based adaptive thresholding for image segmentation in OCR and license plate recognition. — https://www.pirahansiah.com/notes/pubs/journals/adaptive-thresholding/\n\n\nAdaptive_Image_Thresholding_Based_on_the_Peak_Signal-to-noise_Ratio\n\nspotify\n\nPDF Download\nPDF Download\n\n\n\nMind Map: Adaptive Image Thresholding Based on PSNR\n\n1. Introduction\n\n Research Focus: Enhancing image thresholding techniques using Peak Signal-to-noise Ratio (PSNR.\n Key Concept: PSNR-based thresholding improves object-background separation in images.\n Applications: License Plate Recognition (LPR, Optical Character Recognition (OCR, Standard and Handwritten Image Processing.\n\n\n2. Key Components\n\n2.1 Thresholding\n\n Definition: Segmentation technique used to differentiate objects from the background in images.\n Types:\n \n Single Thresholding: Produces binary images; faster computation.\n Multilevel Thresholding: Produces gray-scale images; more complex but can capture finer details.\n \n \n Importance: Reduces data size and improves computational efficiency in image processing.\n\n\n2.2 Peak Signal-to-noise Ratio (PSNR\n\n Role: Used as an indicator to segment images by measuring image quality.\n Purpose: Helps in selecting optimal threshold values for image segmentation.\n Formula: PSNR measures similarity between the original image and segmented image.\n\n\n3. Image Thresholding Methods\n\n3.1 Single Thresholding\n\n Focus: Binary image generation, separating objects using one threshold value.\n Advantages: Faster processing and simpler implementatio..." },{ "id": "/notes/docs/llm/agents/", "title": "Mind Map: Orchestrating AI Agents", "url": "/notes/docs/llm/agents/", "tags": ["ai","llm","agents","multi-agent"], "hashtags": "#ai #llm #agents #multiagent", "category": "hub", "body": "\n Mind Map: Orchestrating AI Agents — Mind map and guide to orchestrating multiple AI agents for complex task completion using LLMs. — https://www.pirahansiah.com/notes/docs/llm/agents/\n\n\n\n\nMind Map: Orchestrating Agents\n\n1. Introduction\n\n Concept: Coordination of multiple AI agents to perform complex tasks\n Goal: To achieve tasks that are too complicated for a single agent to handle\n Example Tasks: Research, planning, multi-step processes\n\n\n2. Key Components\n2.1 Agents\n\n Definition: Autonomous units designed to carry out specific tasks\n Types:\n \n Single-purpose agents\n General-purpose agents\n \n \n Capabilities:\n \n Interact with environments\n Process inputs and produce outputs\n Self-contained decision-making\n \n \n\n\n2.2 Orchestrator\n\n Role: Coordinates and manages multiple agents\n Tasks:\n \n Delegates tasks among agents\n Monitors agent progress\n Handles communication between agents\n Combines results from various agents to complete the overall task\n \n \n\n\n2.3 Communication\n\n Importance: Enables agents to work together\n Methods:\n \n Message passing between agents\n API calls between different AI models or functions\n Shared memory or database for information exchange\n \n \n\n\n3. Orchestrating Multiple Agents\n3.1 Task Decomposition\n\n Purpose: Breaking down complex tasks into manageable subtasks\n Method:\n \n Assign subtasks to specialized agents\n Monitor each agent’s progress\n Aggregate results from agents\n \n \n\n\n3.2 Decision-Making\n\n Orchestrator Role:\n \n S..." },{ "id": "/notes/docs/projects/ai-cost-calculator/", "title": "AI Model Cost Calculator", "url": "/notes/docs/projects/ai-cost-calculator/", "tags": ["ai","llm","cost","calculator","gpt4","claude","gemini"], "hashtags": "#ai #llm #costcalculator #gpt4 #claude #gemini", "category": "hub", "body": "\n AI Model Cost Calculator — Estimate text and image processing costs for GPT-4, Claude, and Gemini — token-aware cost planning for computer vision and multimodal AI. — https://www.pirahansiah.com/notes/docs/projects/ai-cost-calculator/\n\n\nAI Model Cost Calculator\n\nEstimate token and processing costs for multimodal AI models — GPT-4 Turbo with Vision, Claude, and Google Gemini. Use this guide to budget computer-vision and LLM workloads before you build.\n\nWhy Track Token Costs?\n\nVision and language models charge per token. Images can be expensive: a 1080×1080 image in GPT-4 Vision high mode is split into 512-pixel tiles and billed as base cost plus per-tile cost. Understanding the math prevents surprise bills in production.\n\nCost Factors\n\n\n Input tokens — text prompt plus image tokens.\n Output tokens — generated text (usually pricier per token.\n Image mode — low mode (fixed 85 tokens/image vs high mode (base 85 + 170 tokens per 512px tile.\n Model tier — GPT-4 class, Claude Opus/Sonnet, Gemini Pro — each has different rates.\n Caching & batching — prompt caching and batched inference reduce effective cost.\n\n\nHow to Estimate\n\n\n Count prompt tokens (a rough rule: ~1.3 tokens per word, plus image tokens.\n Multiply by the model’s input price per million tokens.\n Add estimated output tokens at the output price.\n Multiply by expected request volume per month.\n\n\nLinks\n\n\n OpenAI community: how to calculate image tokens in GPT-4 Vision\n Google Ad Settings\n\n" },{ "id": "/notes/courses/ai-hardware/", "title": "AI Hardware Accelerators", "url": "/notes/courses/ai-hardware/", "tags": ["AI-hardware","accelerators","NPU","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "\n AI Hardware Accelerators — Workshop covering custom accelerators, NPUs, and edge AI chips. — https://www.pirahansiah.com/notes/courses/ai-hardware/\n\n\nAI Hardware Accelerators\n\nWorkshop on custom AI accelerators, NPUs, and edge computing chips.\n\n#AIHardware #Accelerators #FarshidPirahansiah\n" },{ "id": "/notes/pubs/journals/ant-colony/", "title": "Ant Colony Optimization for Image Processing", "url": "/notes/pubs/journals/ant-colony/", "tags": ["ant-colony-optimization","image-processing","thresholding","optimization"], "hashtags": "#aco #imageprocessing #thresholding #optimization", "category": "hub", "body": "\n Ant Colony Optimization for Image Processing — ACO-based optimization for image thresholding and edge detection in OCR systems. — https://www.pirahansiah.com/notes/pubs/journals/ant-colony/\n\n\nUSING_AN_ANT_COLONY_OPTIMIZATION_ALGORITHM\n\n\n \n https://www.pirahansiah.com/notes/pubs/journals/USING_AN_ANT_COLONY_OPTIMIZATION_ALGORITHM/\n \n \n spotify\n \n \n PDF Download\n\n \n \n\n\nAnt Colony Optimization for Image Edge Detection\n\n1. Introduction\n\n Thresholding: Used in various computer vision applications like OCR, image segmentation, and object tracking.\n Ant Colony Optimization (ACO: Population-based metaheuristic for optimization.\n Objective: Combining ACO, edge detection, and thresholding for Optical Character Recognition (OCR systems.\n\n\n2. State of the Art\n2.1 Thresholding Methods\n\n Categories: Single, Multilevel, Multi-thresholding\n Single Thresholding: Converts the image into binary (black and white.\n \n Pirahansiah’s Single Threshold Method: A custom single threshold method using PSNR.\n \n \n Multilevel Thresholding: Separates objects based on gray values using multiple thresholds.\n Multi-threshold: Uses multiple threshold values to identify objects in images.\n\n\n2.2 Ant Colony Optimization (ACO\n\n Introduction: Initially proposed by Marco Dorigo in 1992 for combinatorial optimization problems.\n Application: Used for image edge detection in this paper.\n Process:\n \n Initialize ants randomly.\n Move ants based on probability and pheromone updates.\n Update pheromone values for optimization.\n \n \n\n\n3. Proposed Method\n\n Com..." },{ "id": "/notes/docs/cuda/apple-ml/", "title": "MLX, CoreML, and Metal for Apple Silicon", "url": "/notes/docs/cuda/apple-ml/", "tags": ["apple-silicon","coreml","metal","mlx"], "hashtags": "#applesilicon #coreml #metal #mlx", "category": "hub", "body": "\n MLX, CoreML, and Metal for Apple Silicon — Guide to Numba JIT, PyCUDA-to-Metal transition, CoreML conversion, and MLX framework on Apple Silicon. — https://www.pirahansiah.com/notes/docs/cuda/apple-ml/\n\n\nNumba JIT Tutorial and PyCUDA with Apple Silicon Adaptation\n\nNumba JIT on Apple Silicon\nThis tutorial explores using Numba’s @jit(nopython=True decorator to optimize Python code for faster execution. The @jit(nopython=True decorator from Numba compiles Python functions into machine code for improved performance, especially for numerical tasks.\n\nBasic Example: Sum of Squares\nfrom numba import jit\n\n@jit(nopython=True\ndef sum_of_squares(n:\n total = 0\n for i in range(n:\n total += i * i\n return total\n\nprint(sum_of_squares(10 # Output: 285\n\n\nOn Apple Silicon, Numba can be used to optimize CPU-bound tasks. Although it doesn’t directly support GPU via Metal or NPU, you can use it to significantly speed up CPU computations, which Apple’s M1/M2 chips handle efficiently with multiple cores.\n\nTransition to Metal for GPU\nTo offload heavy parallel tasks to the GPU, Apple uses Metal, an API for high-performance graphics and computation on macOS. Metal’s Metal Shading Language (MSL provides a way to run GPU tasks that would otherwise be written for CUDA in environments like PyCUDA.\n\nPyCUDA to Metal for Apple Silicon\nPyCUDA is typically used for running GPU tasks on NVIDIA hardware using CUDA. However, on Apple Silicon, you can transition from PyCUDA to Metal for GPU programming. Metal can handle parallel GPU tasks on macOS or iOS.\n\nMetal Shading Language (MSL for Com..." },{ "id": "/notes/docs/llm/avatar/", "title": "Local Video Avatar Generator with Ollama", "url": "/notes/docs/llm/avatar/", "tags": ["ai","llm","video-generation","ollama"], "hashtags": "#ai #llm #videogeneration #ollama", "category": "hub", "body": "\n Local Video Avatar Generator with Ollama — Build a fully local video avatar generator using Ollama LLM, Piper TTS, and Wav2Lip without cloud services. — https://www.pirahansiah.com/notes/docs/llm/avatar/\n\n\nBuilding a Local Video Avatar Generator Using Ollama and Open-Source Tools\n\nCreating a video avatar generator completely locally without cloud services or API keys is challenging but possible. Here’s a step-by-step guide to build a system that generates talking video avatars using locally-run models.\n\nPrerequisites\n\n\n A computer with decent GPU (at least 8GB VRAM recommended\n 16GB+ RAM\n 50GB+ free storage space\n Linux or macOS (Windows with WSL also works\n Basic familiarity with command line\n\n\nStep 1: Set Up Ollama for Local LLM\n\nOllama allows you to run large language models locally for text generation.\n\n\n Install Ollama:\n # For macOS/Linux\ncurl -fsSL https://ollama.com/install.sh sh\n \n# For Windows (via WSL\n# First install WSL, then run the Linux command above\n \n \n Pull a suitable model (Llama3 recommended for better performance:\n ollama pull llama3\n \n \n Test your Ollama installation:\n ollama run llama3 \"Write a short 30-second script about climate change\"\n \n \n\n\nStep 2: Install Local Text-to-Speech Engine\n\nWe’ll use Piper, a fast local TTS system:\n\n\n Install dependencies:\n sudo apt-get update\nsudo apt-get install -y build-essential python3-pip python3-venv\n \n \n Set up a Python virtual environment:\n python3 -m venv ~/venv-tts\nsource ~/venv-tts/bin/activate\n \n \n Install Piper:\n pip install piper-tts\n \n \n Down..." },{ "id": "/notes/docs/llm/blog/", "title": "Blog: AI, LLMs, and Computer Vision", "url": "/notes/docs/llm/blog/", "tags": ["ai","llm","rag","multi-agent"], "hashtags": "#ai #llm #rag #multiagent", "category": "hub", "body": "\n Blog: AI, LLMs, and Computer Vision — Blog posts on RAG vs CAG, multi-agent architectures, AI’s impact on jobs, and the future of software development. — https://www.pirahansiah.com/notes/docs/llm/blog/\n\n\n2025\n\nMay\n\nRAG vs. CAG: Choosing the Right Approach for Your AI Projects\n\n AI Text Generation Methods\n / \\\n / \\\n Generation Approaches Emerging LLM Methods\n / \\ / \\\n / \\ / \\\n RAG CAG Transformer² MML Mosaic\n / \\ / \\ \nAccess to Higher Fast Simple Self- Modular Composite\nup-to-date complex- response architec-adaptive components pruning\n info ity times ture weights \n Better Faster\n reasoning inference\n\n\nUltra-Brief Summary: Compare RAG (retrieval-based, updated info, complex with CAG (cache-based, faster, simpler approaches, plus three new LLM methods: self-adaptive Transformer², modular MML, and efficient Mosaic pruning.\n\nRetrieval-Augmented Generation (RAG\nRAG joins a language model with a retrieval system that gets relevant documents from a knowledge base before creating responses. This works very well with large or frequently updated information sets because it can access the newest information.\n\nAdvantages..." },{ "id": "/notes/courses/book-summary/", "title": "Book Summaries", "url": "/notes/courses/book-summary/", "tags": ["books","summaries","PKM","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "\n Book Summaries — Book summaries and knowledge management resources. — https://www.pirahansiah.com/notes/courses/book-summary/\n\n\nBook Summaries\n\n\n Topics & Projects — All topics and project index\n Machine Learning Specialization — ML fundamentals and course notes\n Full Stack Deep Learning — End-to-end DL deployment\n MLOps — ML pipeline operations and monitoring\n Portfolio & Use Cases — Personal knowledge management and project portfolio\n\n\n#Books #PKM #KnowledgeManagement\n" },{ "id": "/notes/pubs/journals/character-recognition/", "title": "Character & Object Recognition via Features", "url": "/notes/pubs/journals/character-recognition/", "tags": ["character-recognition","object-recognition","feature-extraction","glcm"], "hashtags": "#characterrecognition #objectrecognition #featureextraction #glcm", "category": "hub", "body": "\n Character & Object Recognition via Features — Global feature extraction methods for character and object recognition in computer vision. — https://www.pirahansiah.com/notes/pubs/journals/character-recognition/\n\n\nCHARACTER_AND_OBJECT_RECOGNITION_BASED_ON_GLOBAL_FEATURE_EXTRACTION\n\nspotify\n\nPDF Download\n\n\n\nMind Map: Character and Object Recognition Based on Global Feature Extraction\n\n1. Introduction\n\n Optical Character Recognition (OCR: Recognizes handwritten, irregular, and machine-printed characters.\n Key Tasks in OCR:\n \n Pre-processing\n Segmentation\n Feature Extraction\n Classification\n Recognition\n \n \n\n\n2. Feature Extraction Methods\n2.1 Global Feature Extraction\n\n Definition: Uses entire image characteristics to extract features.\n Methods:\n \n Gray Level Co-occurrence Matrix (GLCM: Uses spatial distribution of gray-level values.\n Edge Direction Matrix (EDMS: Captures edge directions but produces a limited number of features.\n \n \n Challenges:\n \n Less discriminative features.\n Higher dimensionality leads to longer processing times.\n \n \n\n\n2.2 Spatial Feature Extraction\n\n Definition: Focuses on local image characteristics.\n Techniques:\n \n Robinson Compass Mask: Uses gradient filters in eight directions.\n \n \n Strengths: Better for character recognition.\n Limitations: Time-consuming due to high-dimensional data.\n\n\n3. Proposed Method\n\n Combination of GLCM and EDMS:\n \n Aims to improve recognition rates by combining features.\n Feature Selection: Uses gain ratio to reduce fe..." },{ "id": "/notes/courses/cloud-native/", "title": "Cloud-Native with Kubernetes", "url": "/notes/courses/cloud-native/", "tags": ["Docker","Kubernetes","cloud-native","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "\n Cloud-Native with Kubernetes — Docker and Kubernetes fundamentals for cloud-native infrastructure. — https://www.pirahansiah.com/notes/courses/cloud-native/\n\n\nCloud-Native Infrastructure with Kubernetes\n\nDocker Basics\n\ndocker run -ti ubuntu:latest bash\ndocker ps -format $FORMAT\ndocker ps -l\ndocker commit ID\ndocker tag imageID my-image\ndocker run --rm -ti ubuntu sleep 5\ndocker run -d -ti ubuntu bash # detached\ndocker attach name # Ctrl+P /Q to detach\ndocker logs container_name -p\ndocker images\n\n\n#Docker #Kubernetes #CloudNative\n" },{ "id": "/notes/docs/coaching/", "title": "CV Coaching & Teaching Roadmap", "url": "/notes/docs/coaching/", "tags": ["computer-vision","coaching","teaching","curriculum"], "hashtags": "#cv #coaching #teaching #curriculum", "category": "hub", "body": "\n CV Coaching & Teaching Roadmap — Personalized computer vision coaching with a structured curriculum from fundamentals to edge AI. — https://www.pirahansiah.com/notes/docs/coaching/\n\n\nComputer Vision Expertise: Teaching & Coaching Services\n\nWelcome to my professional computer vision teaching service. I offer personalized coaching, tutoring, and online sessions to help you master image processing and computer vision.\n\nAbout Me\n\nI am an experienced computer vision educator with deep expertise across the entire computer vision pipeline. My teaching approach emphasizes building strong foundations while connecting theory to practical applications. I help students develop both theoretical understanding and hands-on implementation skills.\n\nTeaching Philosophy\n\nMy teaching is built around:\n\n Connecting theoretical concepts to real-world applications\n Progressive skill building from fundamentals to advanced topics\n Hands-on projects that reinforce learning\n Personalized guidance based on your background and goals\n\n\nServices Offered\n\n\n One-on-One Tutoring: Personalized sessions tailored to your learning pace and specific interests\n Group Workshops: Collaborative learning environments focused on specific topics\n Project-Based Coaching: Guidance on implementing computer vision in your specific applications\n Code Reviews: Analysis of your implementations with suggestions for improvement\n Career Guidance: Mentorship for those pursuing computer vision careers\n\n\nMy Computer Vision Roadmap\n\nMy comprehensive teaching curriculum follows the roadmap below, which I adapt b..." },{ "id": "/notes/pubs/papers/conference-paper/", "title": "Publications — Conference Papers", "url": "/notes/pubs/papers/conference-paper/", "tags": [], "hashtags": "", "category": "hub", "body": "\n Publications — Conference Papers — All peer-reviewed conference papers by Dr. Farshid Pirahansiah on image processing, camera calibration, pattern recognition, and grid computing. — https://www.pirahansiah.com/notes/pubs/papers/conference-paper/\n\n\ntags: research, papers, image-processing, computer-vision, camera-calibration, pattern-recognition]\nhashtags: “#research #papers #computervision #imageprocessing”\n\nConference Papers\n\nAll peer-reviewed conference papers by Dr. Farshid Pirahansiah.\n\n\n\nImage Processing & Thresholding\n\nAdaptive Image Segmentation Based on PSNR for License Plate Recognition\nAdaptive thresholding using PSNR for license plate recognition systems.\n\nLicense Plate Recognition with Multi-Threshold Based on Entropy\nEntropy-based multi-threshold approach for license plate detection.\n\nMulti-threshold Approach for License Plate Recognition System\nMulti-level thresholding with adaptive MLP for license plate recognition.\n\nComparison of Single Thresholding Method for Handwritten Images Segmentation\nComparison of Otsu and PSNR-based thresholding for handwritten image segmentation.\n\n\n\nCamera Calibration & 3D Vision\n\nCamera Calibration for Multi-Modal Robot Vision\nAutomatic calibration framework for multi-modal robot vision systems.\n\nPattern Image Significance for Camera Calibration\nEvaluation of calibration pattern significance for robot vision.\n\n2D versus 3D Map for Environment Movement Objects\nComparison of 2D and 3D mapping for environment movement detection.\n\n\n\nPattern Recognition & Classification\n\nCharacter Recognition Based on Global Feature\nG..." },{ "id": "/notes/docs/cpp/", "title": "C++ Quick Reference", "url": "/notes/docs/cpp/", "tags": ["cpp","programming","debugging","memory-management"], "hashtags": "#cpp #programming #debugging #memorymanagement", "category": "hub", "body": "\n C++ Quick Reference — Quick reference for C++ memory management, data structures, debugging tools, and shell basics. — https://www.pirahansiah.com/notes/docs/cpp/\n\n\nC++\n\nhash\nstd::map<string,int> prices;\nprices‘aa’]=310;\nprices={‘aa’:310, ‘bb’:410}\n\nstack\nstd::stack stk;\nstk.push(5;\nstk.pop(; //5\n\nqueue\nstd::queue q;\nq.push(5; ... 89\nq.pop(; //5\n\nfrom collections import deque\nstk=deque(\nstk.append(‘asadfas’\nstk.pop(\n\n o(n\nstk=]\nstk.pop(.rstk-1]\n\n\ncollections import lifo\ndeque(\nappendLeft(5\npop\n\nstack\n\n automatically managed\n LIFO\n at compile time\n short term\n 8 MB : 8192 kb\n fast\n ulimit_a : show memory\n push / pop\n\n\nheap\n\n need managed by yourself\n if we need to use more than 8 MB we need to use heap\n run time;\n dynamic\n long time\n new/delete\n slower\n \n pointer\n \n \n \n \n \n pmap ‘pidof _____’\n tail _n1\n grep_o ‘0-9]*’\n awk ‘{print $ 0/(1024*1024” GiB]”}’\n \n \n \n \n tools\n \n valgrind ./my_program\n fsanitize = address\n \n \n echo %errorlevel%\n gflags /i print+Greeting.ext +sls\n \n cdb printGreeting.ext\n \n start with “/” is absolute path\n start with “folder/file…” is relative path\n ”/” is root\n ”~” is home folder\n ”.” is current folder\n \n ”..” is parent folder\n \n a-c] is abc\n grep & ls.txt search inside file\n ”;” calls all command one after\n && same but if error strop next\n \n \n \n \n ”\n ” pipe\n \n \n \n \n htop\n struct like class that all members is public using for simple data\n \n\n\nlog ..." },{ "id": "/notes/docs/cuda/cuda-vscode/", "title": "CUDA Development Environment in VS Code on Windows", "url": "/notes/docs/cuda/cuda-vscode/", "tags": ["cuda","gpu","development","visual-studio-code"], "hashtags": "#cuda #gpu #devtools #vscode", "category": "hub", "body": "\n CUDA Development Environment in VS Code on Windows — Step-by-step guide to configure VS Code for CUDA C++ development on Windows with nvcc, IntelliSense, and debugging. — https://www.pirahansiah.com/notes/docs/cuda/cuda-vscode/\n\n\nSimple Setting Up a CUDA Development Environment in VS Code in Windows\n\nhttps://www.pirahansiah.com/notes/docs/cuda/cuda-vscode/ \n\n\n🚀 Setting Up a CUDA Development Environment in VS Code (Windows\n\nIf you’re working with CUDA C++ and want a clean and efficient workflow inside Visual Studio Code, this guide shows how to configure tasks and launch settings to build and debug .cu files using nvcc and the MSVC toolchain.\n\nBelow is a breakdown of how the key configuration files come together.\n\n⸻\n\n🛠️ tasks.json – Automating the Build Process\n{\n \"version\": \"2.0.0\",\n \"tasks\": \n {\n \"label\": \"Build CUDA Project\",\n \"type\": \"shell\",\n \"command\": \"nvcc\",\n \"args\": \n \"-I\",\n \"C:\\\\Program Files\\\\Microsoft Visual Studio\\\\2022\\\\Enterprise\\\\VC\\\\Tools\\\\MSVC\\\\14.42.34433\\\\include\",\n \"-L\",\n \"C:\\\\Program Files\\\\Microsoft Visual Studio\\\\2022\\\\Enterprise\\\\VC\\\\Tools\\\\MSVC\\\\14.42.34433\\\\lib\\\\x64\",\n \"${workspaceFolder}/main.cu\",\n \"-o\",\n \"${workspaceFolder}/main.exe\"\n ],\n \"group\": {\n \"kind\": \"build\",\n \"isDefault\": true\n },\n \"problemMatcher\": ]\n }\n ]\n}\n\n\nThis task uses nvcc to compile a CUDA file (main.cu and links it against MSVC libraries. It’s defined as the default build task, making it easy to trigger with Ctrl + Shift + B.\n\n⸻\n\n💡 settings.json – File Associati..." },{ "id": "/notes/pubs/books/ai/cv-meets-llm/", "title": "Computer Vision Meets LLM", "url": "/notes/pubs/books/ai/cv-meets-llm/", "tags": [], "hashtags": "", "category": "hub", "body": "\n Computer Vision Meets LLM — Book exploring multimodal AI combining computer vision with large language models. — https://www.pirahansiah.com/notes/pubs/books/ai/cv-meets-llm/\n\n\ntags: computer-vision, llm, multimodal, rag]\nhashtags: “#cv #llm #multimodal #rag”\n\nFarshid Pirahansiah\n\nComputer Vision Meets LLM\n\nAI computer vision locally LLMs on device\n\nComputer Vision Meets LLM: Multi-Agent Swarm with RAG for Images and Videos\n\nIntroduction\n\nThe convergence of Computer Vision (CV and Large Language Models (LLMs marks a significant advancement in artificial intelligence, enabling more comprehensive and intelligent systems capable of understanding and interacting with the world in multimodal ways. By integrating multi-agent swarms with Retrieval-Augmented Generation (RAG, developers can create sophisticated applications that process and analyze images and videos alongside textual data. This synergy enhances capabilities in areas such as image recognition, video analysis, document processing, and interactive user experiences.\n\n1. Integration of Computer Vision and Large Language Models\n\nCombining CV and LLMs leverages the strengths of both modalities:\n\n Computer Vision excels in interpreting and analyzing visual data, identifying patterns, objects, and actions within images and videos.\n Large Language Models (e.g., GPT-4 are proficient in understanding and generating human-like text, enabling nuanced interactions and contextual understanding.\n\n\nTogether, they enable applications that require both visual and textual comprehension, such as automated content creation, intell..." },{ "id": "/notes/docs/cv-overview/", "title": "Dr. Farshid Pirahansiah — CV & AI", "url": "/notes/docs/cv-overview/", "tags": ["computer-vision","ai","portfolio","about"], "hashtags": "#cv #ai #portfolio #about", "category": "hub", "body": "\n Dr. Farshid Pirahansiah — CV & AI — Dr. Farshid Pirahansiah — CV & AI — https://www.pirahansiah.com/notes/docs/cv-overview/\n\n\nDr. Farshid Pirahansiah\n\n\n \n \n Computer Vision Research Engineer & Technical Lead\n Berlin, Germany\n \n \n\n\nAbout Me\n\nI am an accomplished Research Engineer with 12+ years of experience, including a PhD in Computer Science. My career has been dedicated to Computer Vision, Machine Learning, and ML Operations, with a proven track record of transforming complex algorithms into production-ready applications.\n\nExperience\n\n\n 12+ years: Computer Vision, C++, R&D\n 10+ years: Machine Learning, Deep Learning, Python, Embedded Systems, Multi-Camera Systems\n 7+ years: IoT, Model Optimization, Robotics, Medical Imaging, Cloud (AWS\n 5+ years: Technical Lead, Global Collaboration\n 2+ years: LLMs, Multimodal AI, RAG, Agentic Workflows\n\n\nCore Skills\n\n\n Computer Vision & AI: Image processing, deep learning, real-time systems\n Languages: Python, C++, MATLAB\n Frameworks: PyTorch, TensorFlow, ONNX Runtime, TensorRT, OpenVINO\n Tools: Docker, Kubernetes, AWS, Git, MLflow, CI/CD\n Edge AI: Jetson, Coral TPU, model quantization (INT8/FP16\n\n\nPublished Research\n\n\n 3 Patents (Face Image Augmentation, Vehicle Detection, Facial Analysis Advertising\n 2 Book Chapters (Springer\n \n \n \n \n 6 Journal Papers\n 11 Conference Papers\n \n \n \n \n Full Portfolio\n\n\nConsulting Services\n\nI offer personalized coaching, tutoring, and consulting in computer vision and AI. Learn more.\n\n\n\nContent Hub\n\nC..." },{ "id": "/notes/pubs/cv/", "title": "CV — Dr. Farshid Pirahansiah", "url": "/notes/pubs/cv/", "tags": ["computer-vision","resume","publications","patents"], "hashtags": "#cv #resume #publications #patents", "category": "hub", "body": "\n CV — Dr. Farshid Pirahansiah — Curriculum vitae covering 12+ years of computer vision R&D, 21 publications, and 3 patents. — https://www.pirahansiah.com/notes/pubs/cv/\n\n\nAs an AI and computer vision expert with over a decade of experience collaborating with esteemed global organizations, my expertise encompasses AI research and development, machine learning, deep learning, Internet of Things (IoT, and model optimization for edge and cloud-based solutions. With a portfolio of 21 publications, three patents, and extensive practical experience in real-time computer vision applications, I have spearheaded groundbreaking projects in generative AI, video analytics, and intelligent systems. Proficient in C++, Python, OpenCV, and advanced GPU optimization, I am recognized for bridging the gap between cutting-edge research and commercially viable products.\n\nMe\n\n I have a total of 21 publications, including 3 patents, 2 book chapters, 6 journal articles, and 10 conference papers. My Google Scholar profile indicates 137 citations, an h-index of 7, and an i10-index of 5.\n Dr. Farshid Pirahansiah CV\n My podcast\n\n\nMy Academic Contributions and Publications\n\n \n My Google Scholar Profile\n\n \n \n \n Metric\n Value\n \n \n \n \n Citations\n 142\n \n \n h-index\n 8\n \n \n i10-index\n 5\n \n \n \n \n ScienceOpen Collection\n My Publications ( Total 21 \n My Patents (3\n \n My Patents: A METHOD FOR AUGMENTING A PLURALITY OF FACE IMAGES ..." },{ "id": "/notes/docs/dev-tools/", "title": "Developer Tools & Setup Guide", "url": "/notes/docs/dev-tools/", "tags": ["developer-tools","docker","github","productivity"], "hashtags": "#devtools #docker #github #productivity", "category": "hub", "body": "\n Developer Tools & Setup Guide — Curated developer tools, shell essentials, Docker tips, and GitHub tricks. — https://www.pirahansiah.com/notes/docs/dev-tools/\n\n\nRecommended Tools\n\n\n NeoHtop — Modern htop alternative\n Cap — Open source Loom alternative for screen recordings\n RustDesk — Open source remote desktop\n Helix Editor — Post-modern modal editor\n Blender MCP — Blender + MCP integration\n RX Resume — Resume builder\n Practical Computer Vision — CV learning resources\n CUDA Codes — CUDA flash attention algorithms\n Intel RealSense HDR — HDR depth cameras\n Apple ML-GBC — Apple ML framework\n Python CLI — Python CLI tool\n Data Structures for Image Processing — Newsletter\n\n\n\n\nShell Essentials\n\nEcho & Variables\n\n echo \"value is $foo\" → value is bar (variable expanded\n echo 'value is $foo' → value is $foo (literal\n foo=bar — no space around =\n $PATH — environment variable\n cd - — go to previous directory\n\n\nNavigation\n\n / — root directory\n ~ — home folder\n . — current folder\n .. — parent folder\n Absolute path starts with /, relative path starts with folder/file\n\n\nShortcuts\n\n Ctrl+L — clear terminal\n Ctrl+R — reverse search history\n !! — repeat last command\n\n\nPipes & Redirection\n\n ; — run commands sequentially\n && — run next only if previous succeeded\n — pipe output to next command\n >> — append to file\n # — root/sudo prompt\n\n\nUseful Commands\n\n xdg-open file — open with default app\n ulimit -a — show memory limits\n htop — process monitor\n tldr — simplified man pages\n locate / ripgrep (rg / fzf / broot / nnn — search..." },{ "id": "/notes/courses/edge-ai-summit/", "title": "Edge AI Summit 2020", "url": "/notes/courses/edge-ai-summit/", "tags": ["edge-AI","summit","tinyML","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "\n Edge AI Summit 2020 — Notes from Edge AI Summit covering tinyML, neuromorphic computing, and edge inference. — https://www.pirahansiah.com/notes/courses/edge-ai-summit/\n\n\nEdge AI Summit 2020 (Nov 18-20\n\nWednesday Nov 18\n\n Predictive Maintenance at Sensor Level (SensiML\n Fish Farming with Deep Learning (Aquabyte\n TinyMLPerf: Benchmarking Ultra-low Power ML\n Neuromorphic Intelligence for Sensor Edge (Innatera\n Breaking Barriers to Deploy DNNs on Low-Power Hardware (Deeplite\n Optimizing ML Models at Edge (OctoML/Apache TVM\n\n\nThursday Nov 19\n\n Edge AI for Post-Pandemic Society (FogHorn\n Evolving Landscape of Edge AI (Google Coral\n InferX X1 Edge Inference Accelerator (Flex Logix\n Edge in Retail: Walmart Case Study (NVIDIA\n Analog AI Compute (Mythic\n Edge AI for Repetitive Motion (Bosch Sensortec\n\n\nFriday Nov 20\n\n Spatial Computing: Edge + Cloud (Magic Leap\n Autonomous IoT Network (Rakuten Mobile\n Practical Edge Inferencing with Sparsity (GrAI Matter Labs\n Large Scale DL on Edge\n AI Hardware Accelerators 2020-21 (Kisaco Research\n\n\n#EdgeAI #TinyML #Summit #FarshidPirahansiah\n" },{ "id": "/notes/pubs/patents/face-augmentation/", "title": "Face Image Augmentation WO2021060971A1", "url": "/notes/pubs/patents/face-augmentation/", "tags": ["patent","face-augmentation","gan","data-augmentation"], "hashtags": "#patent #faceaugmentation #gan #dataaugmentation", "category": "hub", "body": "\n Face Image Augmentation WO2021060971A1 — GAN-based method for augmenting face images from surveillance with fuzzy logic quality filtering. — https://www.pirahansiah.com/notes/pubs/patents/face-augmentation/\n\n\nspotify\n\n\n \n Your browser does not support the audio element.\n\n\nPDF Download\n\n\n\nThe patent WO2021060971A1 describes a method for augmenting face images, particularly for use in video surveillance systems. The invention addresses the limitations of traditional surveillance, where cameras capture face images from limited angles, leading to incomplete or substandard images. The method involves acquiring face images from both cameras and the internet, applying data augmentation techniques to increase the number of images, and using a Generative Adversarial Network (GAN to create additional face images. This process helps generate better-quality face images that can improve facial recognition systems.\n\nThe system includes several components, such as an image acquisition module connected to a camera, data input and augmentation modules, and a GAN module. The data augmentation modules apply transformations like rotations and flips to increase the variety of face images, while the GAN module generates new images based on trained models. A fuzzy logic module is also employed to evaluate the quality of the generated images, ensuring that only the best images are selected and stored for training a deep learning module. This deep learning module further refines the facial recognition process by extracting higher-level features from the images.\n\nThe method also emphasizes t..." },{ "id": "/notes/pubs/patents/facial-analysis/", "title": "Facial Analysis Advertisement WO2020141969A2", "url": "/notes/pubs/patents/facial-analysis/", "tags": [], "hashtags": "", "category": "hub", "body": "\n Facial Analysis Advertisement WO2020141969A2 — System for targeted digital advertising using real-time facial analysis and demographic detection. — https://www.pirahansiah.com/notes/pubs/patents/facial-analysis/\n\n\ndate_modified: 2024-10-12\ntags: patent, facial-recognition, advertising, deep-learning]\nhashtags: “#patent #facialrecognition #advertising #deeplearning”\n—\n\nspotify\n\n\n \n Your browser does not support the audio element.\n\n\nPDF Download\n\n\n\n🎧 Listen and Explore 🗺️ the engaging audio discussion about the innovative patent “SYSTEM AND METHOD FOR PROVIDING ADVERTISEMENT CONTENTS BASED ON FACIAL ANALYSIS WO2020141969A2”!\n\nThis system leverages facial recognition technology to deliver personalized, dynamic ads tailored to users based on age, gender, emotions, and more—while prioritizing privacy! 🔒✨ Discover how AI is revolutionizing digital advertising by matching ads to user profiles without the need for direct personal data collection.\n\n🔗 Mindmap visualization available too! Uncover how this innovation is shaping the future of advertising. #AI #FacialRecognition #DigitalMarketing #Innovation #LLM\n\nSYSTEM AND METHOD FOR PROVIDING ADVERTISEMENT CONTENTS BASED ON FACIAL ANALYSIS WO2020141969A2\n\nIn this text delves into his patented innovation that transforms digital advertising by utilizing facial analysis to deliver targeted content. The system captures images of users through an embedded camera in digital signage and employs advanced face detection and analysis modules to extract key facial features such as age range, gender, emotions (e.g., happy, sad, angry, sty..." },{ "id": "/notes/courses/fsdl-2022/", "title": "FSDL 2022 Notes", "url": "/notes/courses/fsdl-2022/", "tags": ["FSDL","deep-learning","MLOps","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "\n FSDL 2022 Notes — Notes from Full Stack Deep Learning 2022 covering ML lifecycle, data management, and continual learning. — https://www.pirahansiah.com/notes/courses/fsdl-2022/\n\n\nFull Stack Deep Learning 2022\n\nLecture 01: When to Use ML and Course Vision\n\n Formulating problems and estimating project cost\n Sourcing, cleaning, processing, labeling data\n Picking the right framework and compute\n Troubleshooting training and ensuring reproducibility\n Deploying the model at scale\n Monitoring and continually improving\n\n\nLecture 02: Development Infrastructure & Tooling\n\nLecture 03: Testing\n\nLecture 04: Data Management\n\n Data sources: filesystem, object storage, database, data warehouse, data lake\n SQL and DataFrames (DASK, RAPIDS\n Airflow, Prefect, Dagster\n Feature stores: Tecton, Feast, Featureform\n Labeling: Label Studio, Diffgram, Snorkel.ai\n Data versioning: DVC\n\n\nLecture 06: Continual Learning\n\n Monitoring metrics: data quality, distribution drift\n System monitoring: Datadog, Honeycomb, NewRelic\n Data curation: random, stratified, active learning\n Retraining triggers and dataset formation\n Online testing: shadow mode, A/B testing\n\n\n#FSDL #DeepLearning #MLOps #FarshidPirahansiah\n" },{ "id": "/notes/courses/fsdl/", "title": "Full Stack Deep Learning", "url": "/notes/courses/fsdl/", "tags": ["FSDL","full-stack-deep-learning","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "\n Full Stack Deep Learning — Full Stack Deep Learning course notes covering ML production from data to deployment. — https://www.pirahansiah.com/notes/courses/fsdl/\n\n\nFull Stack Deep Learning\n\nFull Stack Deep Learning (fullstackdeeplearning.com — Notes for week 1 to week 12 (2021\n\nReference: https://fullstackdeeplearning.com/spring2021\n\nUnderfitting vs Overfitting\n\nUnderfitting (reducing bias:\n\n Bigger model\n Reduce regularization\n Error analysis\n Different model architecture\n Tune hyper-parameters\n Add features\n\n\nOverfitting (reducing variance:\n\n Add more training data\n Add normalization (batch norm, layer norm\n Add data augmentation\n Increase regularization (dropout, L2, weight decay\n Error analysis\n Choose different model architecture\n Tune hyper-parameters\n Early stopping\n Remove features\n Reduce model size\n\n\n#StackDeepLearning #computervision #AI #deeplearning\n" },{ "id": "/notes/pubs/journals/fuzzy-thresholding/", "title": "GSFT-PSNR: Fuzzy Threshold Method", "url": "/notes/pubs/journals/fuzzy-thresholding/", "tags": ["fuzzy-logic","thresholding","psnr","image-segmentation"], "hashtags": "#fuzzylogic #thresholding #psnr #imagesegmentation", "category": "hub", "body": "\n GSFT-PSNR: Fuzzy Threshold Method — Fuzzy logic approach to global single thresholding using PSNR for image segmentation. — https://www.pirahansiah.com/notes/pubs/journals/fuzzy-thresholding/\n\n\nGSFT-PSNR_Global_Single_Fuzzy_Threshold\n\nspotify\n\nPDF Download\n\n\n\nMind Map: GSFT-PSNR Global Single Fuzzy Threshold\n\n1. Introduction\n\n Thresholding Importance: Critical in OCR, image analysis, and camera calibration.\n Challenges: Handling varying lighting and environments.\n\n\n2. Key Contributions\n\n GSFT-PSNR Method: Combines fuzzy logic and PSNR for adaptive thresholding.\n \n Applications: OCR, license plate recognition, handwritten document processing.\n Strengths: Works well in varied lighting and complex environments.\n \n \n\n\n3. Materials and Methods\n3.1 One Level Thresholding\n\n Single Thresholding: Faster, effective for binary image conversion.\n Equation: Conversion to binary based on a threshold value.\n\n\n3.2 Multilevel Thresholding\n\n More Detailed: Captures more pixel values, but higher computational cost.\n Application: Best for grayscale images with complex scenes.\n\n\n3.3 Proposed Method (GSFT-PSNR\n\n Key Features:\n \n Uses PSNR as an indicator to determine the optimal threshold.\n Refined using fuzzy logic to adapt to lighting changes.\n \n \n Process:\n \n Calculate PSNR for a range of thresholds.\n Apply fuzzy logic to adjust based on the average intensity value (AIV.\n \n \n\n\n4. PSNR Equation\n\n Formula: Measures the mean square error (MSE between original and thresholded images.\n\nPSNR = 10 \\cdot \\log_{10}\\left(\\frac{MAX^2}{MS..." },{ "id": "/notes/pubs/papers/geometrical-topological/", "title": "Classification with Geometrical Features", "url": "/notes/pubs/papers/geometrical-topological/", "tags": ["classification","geometrical-features","pattern-recognition","machine-learning"], "hashtags": "#classification #geometricalfeatures #patternrecognition #ml", "category": "hub", "body": "\n Classification with Geometrical Features — Evaluation of classification techniques using enhanced geometrical and topological features. — https://www.pirahansiah.com/notes/pubs/papers/geometrical-topological/\n\n\nAn_evaluation_of_classification_techniques_using_enhanced_Geometrical_Topological_Feature_Analysis\n\nhttps://www.pirahansiah.com/notes/pubs/papers/An_evaluation_of_classification_techniques_using_enhanced_Geometrical_Topological_Feature_Analysis\n\nspotify\n\nPDF Download My Conference Paper\n\n\n\nAn Evaluation of Classification Techniques Using Enhanced Geometrical Topological Feature Analysis\n\n1. Introduction\n\n Objective: Evaluation of classification techniques for the Malaysian License Plate Recognition (LPR system.\n Applications of LPR:\n \n Law enforcement\n Border protection\n Vehicle theft detection\n Automatic toll collection\n Traffic control\n \n \n\n\n2. Image Classification Techniques\n\n Artificial Immune Recognition System (AIRS:\n \n Mimics biological immune systems for pattern recognition.\n \n \n Neural Networks (NN:\n \n Machine learning technique that models the human brain to classify images.\n \n \n Bayesian Networks (BN:\n \n Probabilistic graphical models to classify data based on probability distributions.\n \n \n Support Vector Machine (SVM:\n \n Uses geometric representations to classify data.\n \n \n\n\n3. Enhanced Geometrical Topological Feature Analysis\n\n Proposed Approach:\n \n Focuses on improving image classification accuracy for Malaysian license plates.\n Uses topological fe..." },{ "id": "/notes/pubs/papers/global-feature-recognition/", "title": "Character Recognition via Global Features", "url": "/notes/pubs/papers/global-feature-recognition/", "tags": ["character-recognition","feature-extraction","glcm","ocr"], "hashtags": "#characterrecognition #featureextraction #glcm #ocr", "category": "hub", "body": "\n Character Recognition via Global Features — Hybrid GLCM and EDMS method for character recognition in OCR applications. — https://www.pirahansiah.com/notes/pubs/papers/global-feature-recognition/\n\n\nCharacter_Recognition_Based_on_Global_Feature\n\nhttps://www.pirahansiah.com/notes/pubs/papers/Character_Recognition_Based_on_Global_Feature\n\nspotify\n\nPDF Download My Conference Paper\n\n\n\nCharacter Recognition Based on Global Feature Extraction\n\n1. Introduction\n\n Objective: Propose a combination of two feature extraction techniques for character recognition.\n Key Techniques:\n \n Gray Level Co-occurrence Matrix (GLCM\n Edge Direction Matrix (EDMS\n \n \n Challenges:\n \n Selecting the best feature extraction technique for various character recognition tasks.\n \n \n\n\n2. Feature Extraction Techniques\n\n Gray Level Co-occurrence Matrix (GLCM:\n \n Focuses on texture and pixel relationships in the image.\n Extracts texture features for character recognition.\n \n \n Edge Direction Matrix (EDMS:\n \n Emphasizes edges and directions in character images.\n Extracts shape features.\n \n \n\n\n3. Proposed Method\n\n Combination of GLCM and EDMS:\n \n A hybrid approach that leverages both texture and edge features.\n Aims to improve accuracy over using either GLCM or EDMS alone.\n \n \n\n\n4. Classification Techniques\n\n Classifiers Used:\n \n Neural Networks (NN\n Bayesian Networks (BN\n Decision Tree Classifiers\n \n \n Objective: Find the best classifier to complement the hybrid feature extraction method.\n\n\n5. Experime..." },{ "id": "/notes/pubs/papers/grid-computing/", "title": "TafreshGrid: Grid Computing", "url": "/notes/pubs/papers/grid-computing/", "tags": ["grid-computing","distributed-systems","high-performance-computing"], "hashtags": "#gridcomputing #distributedsystems #hpc", "category": "hub", "body": "\n TafreshGrid: Grid Computing — Grid computing infrastructure implementation and evaluation at Tafresh University. — https://www.pirahansiah.com/notes/pubs/papers/grid-computing/\n\n\nTafreshGrid_Grid_computing_in_Tafresh_university\n\nhttps://www.pirahansiah.com/notes/pubs/papers/TafreshGrid_Grid_computing_in_Tafresh_university\n\nspotify\n\nPDF Download My Conference Paper\n\n\n\nTafreshGrid: Grid Computing in Tafresh University\n\n1. Introduction\n\n Objective: Introduce TafreshGrid as a grid computing project aimed at solving computational problems in multiple institutions.\n Grid Computing:\n \n Coordinates distributed computing resources.\n Makes a virtual single image of heterogeneous resources to solve complex problems.\n \n \n\n\n2. Project Overview\n\n Phases of TafreshGrid Project:\n \n Implementation: Setting up the grid infrastructure at Azad University, Tafresh branch.\n Development: Extending the project to all Azad University branches.\n Expansion: Connecting TafreshGrid to the national grid computing infrastructure.\n \n \n\n\n3. Pre-Requisites for Grid Computing\n\n Required Infrastructure:\n \n Networking infrastructure (wired/wireless.\n Distributed computational resources.\n Continuous resource-sharing requirements.\n \n \n Software Tools:\n \n Globus Toolkit for managing grid computing tasks.\n Additional tools like J2SDK, APACHE ANT, and Perl for system configuration.\n \n \n\n\n4. Software Installation\n\n Key Tools:\n \n Globus Toolkit 5.0.1 for grid computing.\n Linux (Debian installed on connected computer..." },{ "id": "/notes/tools/image-in-md/", "title": "test mindmap", "url": "/notes/tools/image-in-md/", "tags": [], "hashtags": "", "category": "hub", "body": "\n test mindmap — test mindmap — https://www.pirahansiah.com/notes/tools/image-in-md/\n\n\nreference\ntopic\nimage\n\n\n\n# inside mindmap test\n# level1\n- in a\n - in b\n - in c\n - in d\n - test 4\n\n\n\n\n\n\n\n# level a1\n## level a2\n### level a3\n\n" },{ "id": "/notes/docs/python/", "title": "Python Configuration & C++ Integration", "url": "/notes/docs/python/", "tags": ["python","cpp","configuration","development"], "hashtags": "#python #cpp #configuration #devtools", "category": "hub", "body": "\n Python Configuration & C++ Integration — Python configuration methods comparison plus pybind11 and Cython tips for C++ integration. — https://www.pirahansiah.com/notes/docs/python/\n\n\nPython\nPython\nA comparison of built-in and third-party configuration options for Python projects\nThis guide compares Python’s native configuration methods and popular third-party libraries, helping developers choose the best fit for their project’s needs.\n#Python #DevTools #Configuration #ConfigFiles #PythonTips #OpenSource #MachineLearning #WebDev #FastAP\n\n\n \n Python Configuration Management\n\n \n\n Python Configuration Management\n \n \n Tips and tricks python scale up projects\n\n Ttips\n \n\n\nPython Configuration Management\n\nA comparison of built-in and third-party configuration options for Python projects\n\n📌 Summary\n\nThis guide compares Python’s native configuration methods and popular third-party libraries, helping developers choose the best fit for their project’s needs.\n\n⸻\n\n🔧 Built-in Configuration Tools\n\nConfigParser (INI Files\n\t•\t📄 Simple structured text files\n\t•\t✅ Built-in\n\t•\t❗ Limitations: String-only, no nesting\n\nargparse (Command Line Arguments\n\t•\t🧰 Used in CLI tools\n\t•\t✅ Built-in\n\t•\t📝 Supports help text, types, defaults\n\nEnvironment Variables (os.environ\n\t•\t🔒 Ideal for secrets and deployment\n\t•\t✅ Built-in\n\t•\t❗ Flat and string-only\n\nPython Module as Config\n\t•\t🐍 Python file for configuration\n\t•\t✅ Built-in\n\t•\t🚀 Full flexibility and dynamic logic\n\nJSON Files\n\t•\t📦 Structured data format\n\t•\t✅ Built-in\n\t•\t❗ No comments, strict format\n\nTOML (e.g. pyproject.toml\n\t•\t🧪 Use..." },{ "id": "/notes/docs/llm/", "title": "AI & LLMs", "url": "/notes/docs/llm/", "tags": ["ai","llm","rag","agents"], "hashtags": "#ai #llm #rag #agents", "category": "hub", "body": "\n AI & LLMs — Articles on LLMs, RAG, multi-agent systems, and local AI integration. — https://www.pirahansiah.com/notes/docs/llm/\n\n\nAI & LLMs\n\nArticles on large language models, retrieval-augmented generation, and AI agents for computer vision workflows.\n\nArticles\n\n\n Advanced LLM Concepts — Transformer architecture, attention mechanisms, scaling laws.\n Orchestrating AI Agents — Multi-agent workflows, tool use, RAG pipelines.\n Blog: AI & LLMs — RAG vs CAG, multi-agent architectures, AI’s impact.\n Avatar Generator — Local video avatar with Ollama + Wav2Lip.\n\n" },{ "id": "/notes/docs/projects/solutions/", "title": "Solutions", "url": "/notes/docs/projects/solutions/", "tags": ["portfolio","computer-vision","ai","projects"], "hashtags": "#portfolio #cv #ai #projects", "category": "hub", "body": "\n Solutions — Showcase of industry applications, portfolio projects, and technical skills in computer vision and AI. — https://www.pirahansiah.com/notes/docs/projects/solutions/\n\n\nImpact Portfolio\n" },{ "id": "/notes/docs/prompts/", "title": "Prompt Engineering Templates", "url": "/notes/docs/prompts/", "tags": ["prompt-engineering","llm","ai","nlp"], "hashtags": "#promptengineering #llm #ai #nlp", "category": "hub", "body": "\n Prompt Engineering Templates — Reusable prompt templates for text simplification, mind maps, and structured content processing. — https://www.pirahansiah.com/notes/docs/prompts/\n\n\n\n \n a\nConsider yourself an expert software developer with extended knowledge about C++\n \n \n b\nFor any text I provide, please process it according to these specific guidelines:\n \n\n\nLanguage Requirements\n\n Convert all content to simple English at B1 level (suitable for intermediate English learners\n Use short, clear sentences with basic vocabulary\n Break complex ideas into digestible points\n Aim for maximum clarity and readability\n\n\nRequired Outputs (in this order\n\n Mind Map Visualization: Create a markdown-based mind map showing the key concepts and their relationships\n Ultra-Brief Summary: Provide a concise overview in under 200 characters\n Reorganized Full Text: Present the complete content in a better structured format while preserving all original information\n\n\nSpecific Modifications\n\n Use formatting (bold, italics, headings to enhance readability\n Add bullet points for lists and sequential information\n Insert subheadings to organize longer sections\n\n\nAdditional Guidelines\n\n Maintain academic integrity while simplifying language\n Keep the original meaning intact despite simplification\n Add clarifying notes for culturally-specific concepts\n\n" },{ "id": "/notes/docs/cv/", "title": "Computer Vision", "url": "/notes/docs/cv/", "tags": ["computer-vision","3d","optical-flow"], "hashtags": "#cv #3d #opticalflow", "category": "hub", "body": "\n Computer Vision — Computer vision topics: 3D vision, optical flow, multi-camera systems. — https://www.pirahansiah.com/notes/docs/cv/\n\n\nComputer Vision\n\nGuides and tutorials on computer vision topics.\n\nTopics\n\n\n 3D Vision & Multi-Camera — Stereo vision, depth estimation, synchronized multi-camera pipelines.\n Optical Flow — Dense and sparse optical flow algorithms for motion analysis.\n Multi-Camera Systems — Scaling from 2 to 100+ cameras with GStreamer and GPU acceleration.\n\n" },{ "id": "/notes/docs/optimization/", "title": "CV, DL & ML Optimization Guide", "url": "/notes/docs/optimization/", "tags": ["optimization","deep-learning","machine-learning","computer-vision"], "hashtags": "#optimization #deeplearning #machinelearning #cv", "category": "hub", "body": "\n CV, DL & ML Optimization Guide — Comprehensive guide to optimizing deep learning models, CV systems, and data pipelines. — https://www.pirahansiah.com/notes/docs/optimization/\n\n\nOptimization\n\nDL\n\n1. Model Optimization\n\n Quantization\n \n Convert to lower precision (INT8, FP16\n \n \n Pruning\n \n Remove unnecessary weights or layers\n \n \n Knowledge Distillation\n \n Use smaller “student” models for efficiency\n \n \n\n\n2. Hardware Utilization\n\n GPU/TPU Acceleration\n \n Fully utilize GPUs or TPUs\n Parallelize across multiple devices\n \n \n CUDA and cuDNN\n \n Optimize using CUDA and cuDNN libraries\n \n \n\n\n3. Efficient Data Loading\n\n Multi-threaded Data Loading\n \n Use PyTorch’s DataLoader\n \n \n Real-time Data Augmentation\n \n Perform on-the-fly augmentations\n \n \n\n\n4. Batch Size Tuning\n\n Increase Batch Size\n \n Improves throughput, balancing memory usage\n \n \n\n\n5. Algorithmic Improvements\n\n Early Stopping\n \n Stop training early when performance stabilizes\n \n \n Gradient Checkpointing\n \n Recompute intermediate activations to save memory\n \n \n\n\n6. Efficient Architectures\n\n MobileNet, EfficientNet, ResNet\n \n Architectures optimized for speed and performance\n \n \n\n\n7. Parallelization & Distributed Training\n\n Distributed Training\n \n Spread training across multiple machines\n \n \n\n\n8. Inference Optimization\n\n ONNX Runtime\n \n Convert to ONNX format for platform-optimized deployment\n \n \n TensorRT or OpenVINO\n \n Use Tens..." },{ "id": "/notes/docs/seo/", "title": "SEO for LLM-Powered Search", "url": "/notes/docs/seo/", "tags": ["seo","llm","ai","web-optimization"], "hashtags": "#seo #llm #ai #weboptimization", "category": "hub", "body": "\n SEO for LLM-Powered Search — How to optimize websites for LLM-powered search engines using structured data and AI-friendly content. — https://www.pirahansiah.com/notes/docs/seo/\n\n\nThe New Era of SEO: Optimizing Websites for LLMs\n\nIn today’s rapidly evolving digital landscape, search engine optimization (SEO is undergoing a fundamental transformation with the rise of Large Language Models (LLMs like ChatGPT, Claude, and Gemini. These sophisticated AI systems are now integrated into search experiences, creating a new frontier for website visibility and discoverability.\n\nHow LLMs Are Changing SEO\n\nLLMs don’t just understand keywords—they comprehend context, semantics, and user intent at unprecedented levels. This shift requires website owners to adapt their SEO strategies beyond traditional keyword stuffing and backlink building.\n\nKey Strategies for LLM-Optimized SEO\n\n1. Structured Data & Machine-Readable Content\n\nLLMs thrive on well-organized, structured information. Implementing JSON-LD schema markup (FAQ, HowTo, Article, Product helps these models understand and extract information from your content efficiently. For example:\n\n{\n \"@context\": \"https://schema.org\",\n \"@type\": \"FAQPage\",\n \"mainEntity\": \n {\n \"@type\": \"Question\",\n \"name\": \"How do I optimize my website for AI search?\",\n \"acceptedAnswer\": {\n \"@type\": \"Answer\",\n \"text\": \"To optimize for AI search, use structured data, clear headlines, and authoritative sources.\"\n }\n }\n ]\n}\n\n\nThis structured approach increases the likelihood that LLMs will reference your website wh..." },{ "id": "/notes/docs/", "title": "Content Hub", "url": "/notes/docs/", "tags": ["computer-vision","ai","llm","edge-ai"], "hashtags": "#cv #ai #llm #edgeai", "category": "hub", "body": "\n Content Hub — Technical articles and tutorials on computer vision, AI, edge deployment, and software engineering. — https://www.pirahansiah.com/notes/docs/\n\n\nContent Hub\n\nTechnical articles, tutorials, and deep dives on computer vision, AI, edge deployment, and software engineering.\n\nComputer Vision\n\n 3D Vision & Multi-Camera Systems — Stereo vision, depth estimation, and synchronized multi-camera pipelines.\n Optical Flow — Dense and sparse optical flow algorithms for motion analysis.\n Multi-Camera Systems — Scaling from 2 to 100+ cameras with GStreamer and GPU acceleration.\n CV Coaching Roadmap — Structured learning path from fundamentals to production systems.\n Computer Vision Overview — CV overview and research.\n\n\nCUDA & GPU Computing\n\n Numba JIT Compilation — Accelerate Python loops with just-in-time compilation.\n PyCUDA Kernels — Write custom CUDA kernels from Python.\n CUDA in VS Code — Dev environment setup on Windows.\n MLX, CoreML & Metal — Apple Silicon ML frameworks.\n\n\nAI & LLMs\n\n Advanced LLM Concepts — Transformer architecture, attention mechanisms, and scaling laws.\n Orchestrating AI Agents — Multi-agent workflows, tool use, and RAG pipelines.\n Blog: AI & LLMs — RAG vs CAG, multi-agent architectures, AI’s impact.\n Avatar Generator — Local video avatar with Ollama + Wav2Lip.\n\n\nOptimization\n\n CV/DL/ML Optimization — Model quantization, pruning, and edge deployment strategies.\n Prompt Engineering — Reusable LLM prompt templates.\n\n\nProgramming\n\n C++ Quick Reference — Memory (stack/heap, STL, debugging, Linux essentials.\n Py..." },{ "id": "/notes/courses/", "title": "Topics & Projects", "url": "/notes/courses/", "tags": ["AI","OpenCV","camera-calibration","DRL","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "\n Topics & Projects — Topics and projects covering AI, OpenCV, camera calibration, deep reinforcement learning, and more. — https://www.pirahansiah.com/notes/courses/\n\n\nTopics & Projects\n\nAI & Machine Learning\n\n Advanced LLM Concepts — RAG, embeddings, multimodal\n Orchestrating AI Agents — Multi-agent systems\n AI Hardware Accelerators — NPU, edge AI chips\n OpenVINO — Intel inference optimization\n Machine Learning Specialization — ML fundamentals\n Full Stack Deep Learning — End-to-end DL\n MLOps — ML engineering for production\n\n\nComputer Vision\n\n 3D Vision & Multi-Camera — Point clouds, depth sensing\n Optical Flow — Motion estimation\n Multi-Camera Systems — Scaling to 100+ cameras\n CV Coaching Roadmap — Learning path\n\n\nGetting Started\n\n Machine Learning Specialization — ML fundamentals\n Full Stack Deep Learning — End-to-end DL\n Developer Tools & Setup — Docker, GitHub, CLI tools\n C++ Quick Reference — Memory, STL, debugging\n\n\nCUDA & GPU\n\n Numba JIT Tutorial — Python performance\n PyCUDA Kernels — CUDA from Python\n CUDA in VS Code — Dev environment\n MLX, CoreML & Metal — Apple Silicon ML\n\n\nCourses & Workshops\n\n FSDL 2022 — Continual learning, data management\n MLOps — ML engineering for production\n ROS — Robot Operating System\n Parallel Programming — Python parallel CV\n Modern C++ — C++29 for image processing\n Cloud-Native — Docker & Kubernetes\n TF Deployment — TF.js, TF Lite, mobile ML\n RISC-V — Open-source processor architecture\n Edge AI Summit — TinyML, neuromorphic, edge inference\n Embedded IoT — Embedded s..." },{ "id": "/notes/pubs/papers/", "title": "Conference Papers", "url": "/notes/pubs/papers/", "tags": ["research","paper","image-processing","computer-vision"], "hashtags": "#research #cv #papers #imageprocessing", "category": "hub", "body": "\n Conference Papers — Peer-reviewed conference papers on image processing, camera calibration, and pattern recognition. — https://www.pirahansiah.com/notes/pubs/papers/\n\n\nConference Papers\n\nAll peer-reviewed conference papers by Dr. Farshid Pirahansiah.\n\n\n\nImage Processing & Thresholding\n\nAdaptive Image Segmentation Based on PSNR for License Plate Recognition\nAdaptive thresholding using PSNR for license plate recognition systems.\n\nLicense Plate Recognition with Multi-Threshold Based on Entropy\nEntropy-based multi-threshold approach for license plate detection.\n\nMulti-threshold Approach for License Plate Recognition System\nMulti-level thresholding with adaptive MLP for license plate recognition.\n\nComparison of Single Thresholding Method for Handwritten Images Segmentation\nComparison of Otsu and PSNR-based thresholding for handwritten image segmentation.\n\n\n\nCamera Calibration & 3D Vision\n\nCamera Calibration for Multi-Modal Robot Vision\nAutomatic calibration framework for multi-modal robot vision systems.\n\nPattern Image Significance for Camera Calibration\nEvaluation of calibration pattern significance for robot vision.\n\n2D versus 3D Map for Environment Movement Objects\nComparison of 2D and 3D mapping for environment movement detection.\n\n\n\nPattern Recognition & Classification\n\nCharacter Recognition Based on Global Feature\nGLCM and EDMS-based hybrid method for character recognition.\n\nClassification Techniques Using Enhanced Geometrical Topological Feature Analysis\nEnhanced geometrical and topological features for pattern classification.\n\n\n\nGrid Computing\n\nTafresh..." },{ "id": "/notes/pubs/books/", "title": "Books & Book Chapters", "url": "/notes/pubs/books/", "tags": ["book","opencv","computer-vision","ai"], "hashtags": "#book #opencv #computervision #ai", "category": "hub", "body": "\n Books & Book Chapters — Books and book chapters on computer vision, camera calibration, and AI. — https://www.pirahansiah.com/notes/pubs/books/\n\n\nBooks & Book Chapters\n\nBooks and book chapters authored or co-authored by Dr. Farshid Pirahansiah.\n\n\n\nSpringer Book Chapters\n\nCamera Calibration and Video Stabilization Framework for Robot Localization\nCamera calibration framework for robot localization in “Control Engineering in Robotics and Industrial Automation” (Springer.\n!camera-calibration-video-stabilization.png]]\nComputational Intelligence: Augmented Optical Flow for Video Stabilization\nOptical flow methods for video stabilization in “Computational Intelligence: From Theory to Application”.\n\n\ndrafts\n\nOpenCV 5 Ebook\n\nComputer Vision Meets LLM\nExploring the intersection of computer vision and large language models.\n\nOpenCV 5 — Chapter 1: Image Basics\nImage processing fundamentals: pixel operations, color spaces, histograms.\n\nOpenCV 5 — Chapter 2: Feature Detection\nFeature detection methods: SIFT, SURF, ORB, keypoint matching.\n\nOpenCV 5 — Chapter 3: Advanced Topics\nAdvanced computer vision techniques using OpenCV 5.\n\nLIST \nFROM #]]\n\n\n\n" },{ "id": "/notes/pubs/patents/", "title": "Patents", "url": "/notes/pubs/patents/", "tags": ["patent","face-detection","vehicle-detection","augmented-reality"], "hashtags": "#patent #facedetection #vehicledetection #augmentedreality", "category": "hub", "body": "\n Patents — AI and computer vision patents covering face augmentation, vehicle detection, and facial analysis. — https://www.pirahansiah.com/notes/pubs/patents/\n\n\nPatents\n\nAI and Computer Vision patents by Dr. Farshid Pirahansiah.\n\n\n\nFace Image Augmentation\n\nA Method for Augmenting a Plurality of Face Images — WO 2021/060971 A1\nGAN-based generation of realistic face images from surveillance video with fuzzy logic quality filtering.\n\n\n\nVehicle Detection\n\nA Method for Detecting a Moving Vehicle — WO 2021/107761 A1\nAdvanced image processing for vehicle detection using illumination enhancement, Sobel edge detection, and geometric noise filtering.\n\n\n\nFacial Analysis & Advertising\n\nSystem and Method for Providing Advertisement Contents Based on Facial Analysis — WO 2020/141969 A2\nFacial recognition system that adjusts digital advertisements based on user demographics and emotions without collecting personal data.\n" },{ "id": "/notes/pubs/keynotes/", "title": "Keynotes", "url": "/notes/pubs/keynotes/", "tags": ["keynote","ai","computer-vision","multimodal"], "hashtags": "#keynote #ai #computervision #multimodal", "category": "hub", "body": "\n Keynotes — Technical keynotes on computer vision, LLMs, and multimodal AI. — https://www.pirahansiah.com/notes/pubs/keynotes/\n\n\nKeynotes\n\nTechnical keynotes and presentations by Dr. Farshid Pirahansiah.\n\n\n\nLLMs Meet Computer Vision\n\nExploring the convergence of Large Language Models and Computer Vision. Covers token economics, multimodal AI, RAG systems, and practical applications in document analysis, video understanding, and IoT/robotics.\n" },{ "id": "/notes/pubs/journals/", "title": "Journal Publications", "url": "/notes/pubs/journals/", "tags": ["journal","image-processing","thresholding","pattern-recognition"], "hashtags": "#research #cv #journals #thresholding", "category": "hub", "body": "\n Journal Publications — Peer-reviewed journal articles on image processing, pattern recognition, and robotics. — https://www.pirahansiah.com/notes/pubs/journals/\n\n\nJournal Publications\n\nAll peer-reviewed journal articles by Dr. Farshid Pirahansiah.\n\n\n\nImage Processing & Thresholding\n\nAdaptive Image Thresholding Based on the Peak Signal-to-Noise Ratio\nPSNR-based adaptive thresholding for image segmentation.\n\nGSFT-PSNR: Global Single Fuzzy Threshold\nFuzzy logic approach to global single thresholding using PSNR.\n\nPeak Signal-to-Noise Ratio Based on Threshold Method for Image Segmentation\nPSNR thresholding method for image segmentation applications.\n\n\n\nPattern Recognition & Feature Extraction\n\nCharacter and Object Recognition Based on Global Feature Extraction\nGlobal feature extraction methods for character and object recognition.\n\n\n\nRobotics & 3D Vision\n\n3D SLAM: Simultaneous Localization and Mapping Trends and Humanoid Robot Linkages\nSurvey of 3D SLAM methods and humanoid robot linkage systems.\n\n\n\nOptimization Algorithms\n\nUsing an Ant Colony Optimization Algorithm for Image Processing\nACO-based optimization for image processing tasks.\n" },{ "id": "/", "title": "Dr. Farshid Pirahansiah ; Embedded Computer Vision & Edge AI", "url": "/", "tags": [], "hashtags": "", "category": "hub", "body": "\n\n\n \n Dr. Farshid PirahanSiah \n \n Embedded Computer Vision& Edge AI at Scale\n \n I help businesses turn computer vision research into production-ready embedded applications.\n From model optimization to multi-camera deployment on NVIDIA Jetson, Raspberry Pi, Hailo, Axelera, and beyond.\n \n \n \n Let's Talk on LinkedIn\n \n\n\n\n 12+Years Experience\n 3Patents\n 17+Publications\n 142Citations\n\n\n\n\n \n 👁\n Embedded Computer Vision\n Real-time object detection, tracking, and classification on edge devices. From prototype to production deployment on Jetson, Coral, Hailo, and Axelera accelerators.\n \n OpenCV\n YOLO\n TensorRT\n ONNX\n \n \n\n \n 📈\n Model Optimization & Quantization\n INT8/FP16 quantization, pruning, and knowledge distillation. Reduce model size 4-10x while maintaining accuracy for real-time inference on resource-constrained hardware.\n \n TensorRT\n OpenVINO\n QDQ\n Pruning\n \n \n\n \n 🎥\n Multi-Camera Systems\n Synchronized multi-camera setups for 3D reconstruction, stereo vision, and surveillance. Scaling from 2 to 100+ cameras with GStreamer and GPU-accelerated pipelines.\n \n GStreamer\n CUDA\n DeepStream\n SLAM\n \n \n\n \n 🤖\n LLM & AI Integration\n Integrate vision-LLMs, RAG pipelines, and multi-agent systems into CV workflows. Local inference with Ollama, or cloud APIs for production-scale AI applications.\n \n LLMs\n RAG\n Ollama\n Agents\n \n \n\n \n ⚙\n MLOps & CI/CD\n End-to-..." },{ "id": "/notes/courses/iot-scholarship/", "title": "IoT Scholarship: Edge AI with OpenVINO", "url": "/notes/courses/iot-scholarship/", "tags": ["IoT","OpenVINO","edge-AI","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "\n IoT Scholarship: Edge AI with OpenVINO — IoT and Edge AI with OpenVINO: face recognition, object detection, pose estimation, and semantic segmentation. — https://www.pirahansiah.com/notes/courses/iot-scholarship/\n\n\nIoT Scholarship Foundation\n\nEdge AI with Intel OpenVINO\n\nKey Models\n\n Face recognition: OpenCV 4.1.1\n Image classification: Deep learning with high accuracy\n Object detection: MobileNet SSD (~5 FPS\n Pose estimation: Very fast and accurate\n Coral TPU: Good performance with modifications\n Intel Movidius Stick 2: Good with OpenCV and Python\n\n\nOpenVINO Pipeline\n\n Pre-trained models from Open Model Zoo\n Model Optimizer: TF/PyTorch/Caffe → IR format\n Inference Engine: Run optimized IR models\n Edge deployment: Input streams, MQTT, web serving\n\n\nModel Types\n\n Classification (yes/no, 1000 classes, 20K ImageNet\n Detection (bounding boxes + classification\n Segmentation (semantic: all same class; instance: separate objects\n Pose estimation\n Text recognition\n GANs\n\n\n#IoT #OpenVINO #EdgeAI #FarshidPirahansiah\n" },{ "id": "/notes/courses/iot/", "title": "Embedded IoT", "url": "/notes/courses/iot/", "tags": ["IoT","embedded","edge-AI","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "\n Embedded IoT — Workshop on embedded IoT systems and edge computing applications. — https://www.pirahansiah.com/notes/courses/iot/\n\n\nEmbedded IoT\n\nWorkshop on embedded IoT systems and edge computing applications.\n\n#IoT #Embedded #FarshidPirahansiah\n" },{ "id": "/notes/pubs/papers/license-plate/", "title": "Entropy-Based Multi-Threshold LPR", "url": "/notes/pubs/papers/license-plate/", "tags": ["license-plate-recognition","entropy","thresholding","image-segmentation"], "hashtags": "#lpr #entropy #thresholding #imagesegmentation", "category": "hub", "body": "\n Entropy-Based Multi-Threshold LPR — Multi-threshold license plate recognition using entropy-based image segmentation. — https://www.pirahansiah.com/notes/pubs/papers/license-plate/\n\n\nLicense_Plate_Recognition_with_Multi-Threshold_Based_on_Entropy\n\nhttps://www.pirahansiah.com/notes/pubs/papers/License_Plate_Recognition_with_Multi-Threshold_Based_on_Entropy\n\nspotify\n\nPDF Download My Conference Paper\n\n\n\nLicense Plate Recognition with Multi-threshold based on Entropy\n\n1. Introduction\n\n Objective: Propose a multi-thresholding method for license plate recognition.\n Importance of Thresholding:\n \n Simplifies image segmentation\n Ensures robustness and accuracy in recognizing license plate characters.\n \n \n Challenges:\n \n Selecting the correct threshold values for better segmentation results.\n \n \n\n\n2. Entropy-based Thresholding\n\n Method:\n \n Based on maximizing the cross entropy between the original image and the segmented image.\n Entropy is treated as a probability distribution of the image histogram.\n \n \n Historical Background:\n \n Originally proposed by Pun and later improved by Kapur for image segmentation.\n Entropy-based thresholding is widely used for bi-level and multi-level thresholding.\n \n \n\n\n3. Proposed Method\n\n Multi-thresholding Based on Maximum Entropy:\n \n Selects several threshold values by maximizing the entropy.\n Integrates partial ranges of the image histogram to achieve better segmentation.\n \n \n Comparison with Other Methods:\n \n Compared to single-thresholding technique..." },{ "id": "/notes/docs/linkedin/", "title": "Top LinkedIn Posts 2024", "url": "/notes/docs/linkedin/", "tags": ["linkedin","camera-calibration","cpp","robotics"], "hashtags": "#linkedin #cameracalibration #cpp #robotics", "category": "hub", "body": "\n Top LinkedIn Posts 2024 — Highlights from 2024 LinkedIn posts covering camera calibration, C++, NVIDIA optimization, and robotics. — https://www.pirahansiah.com/notes/docs/linkedin/\n\n\nOver the past year, my standout posts have featured camera calibration, Python, C++, OpenCV, NVIDIA optimizations, advanced C++ techniques, AI integration in robotics with LLMs, computer vision, and machine learning. Explore more: \nLink\n\nMy Top LinkedIn Posts from the Past Year\n\n\n Camera Calibration\n Camera Calibration\n Optimization Methods Computer Vision\n multi-GPU \n NVIDIA Jetson\n On-Device LLM\n On-Device LLM\n C++ \n C++ \n C++\n My experience Robotics 3D SLAM LLM Vision Multimodal\n\n\n1. Camera Calibration and Subpixel Precision\n\nCamera Calibration Post\n\nIn this post, I delved into advanced methods of camera calibration, focusing on geometric analysis and calibration patterns using tools like MATLAB, Python, and OpenCV. I highlighted a C++ algorithm implemented for high-speed, high-accuracy corner detection within calibration patterns, emphasizing rotation and orientation. The process was further refined using subpixel accuracy and noise reduction techniques, enhancing precision in computer vision applications.\n\n\n\n2. Optimization Methods in Computer Vision\n\nOptimization Methods Post\n\nOptimizing NVIDIA GPUs for deep learning has been a crucial topic, especially in multi-GPU setups. In this post, I explored how leveraging CUDA and cuDNN can lead to high-performance AI applications such as video analytics, face recognition, and smart systems. By optimizing GPU resources, we can ac..." },{ "id": "/notes/docs/links/", "title": "Curated Links & Resources", "url": "/notes/docs/links/", "tags": ["resources","tools","tutorials","references"], "hashtags": "#resources #tools #tutorials #references", "category": "hub", "body": "\n Curated Links & Resources — Hand-picked tools, tutorials, and references for CV, AI, and software engineering. — https://www.pirahansiah.com/notes/docs/links/\n\n\nCurated Links & Resources\n\nHand-picked tools, tutorials, and references for computer vision, AI, and software engineering.\n\nComputer Vision\n\n Vulkan C++ Qt Image Viewer — Image viewer built with Qt6, C++, and Vulkan for real-time rendering.\n\n\nCUDA & GPU\n\n CUDA Python Introduction — Getting started with CUDA from Python.\n CUDA Programming Guide (PDF — Official NVIDIA documentation.\n GPU Architecture Overview — Cornell’s GPU computing resource.\n CUDA Flash Attention Algorithms — Implementation of flash attention in CUDA.\n LeetGPU — GPU programming practice platform.\n CUDA YouTube Playlist — Video tutorials on CUDA programming.\n\n\nPython\n\n Pydoll — Browser automation without WebDriver.\n Hyperdiv — Python GUI framework.\n Python Multithreading Tutorial — Concurrent programming in Python.\n Python Data Structures — Essential data structures explained.\n\n\nAI & LLMs\n\n Multimodal RAG with Granite — IBM’s notebook on building RAG systems.\n MCP Server & Client Webinar — Building Model Context Protocol servers.\n AdaptiveCpp GPU Programming — 68x faster C++ GPU programming.\n\n\nFPGA\n\n Introduction to FPGA Part 11 — RISC-V Softcore Processor — Digi-Key’s FPGA tutorial series.\n\n\nStartup & Business\n\n How to Start a Startup (21 videos — Y Combinator’s startup course.\n VentureKit — Business planning tool.\n\n\nDeveloper Tools\n\n pickhost — Simple SSH host manager.\n Voidtools Everything — Fast W..." },{ "id": "/notes/pkm/links/", "title": "Curated Links & Resources", "url": "/notes/pkm/links/", "tags": ["curated-links","ai-tools","resources"], "hashtags": "#curatedlinks #aitools #resources", "category": "pkm", "body": "\n Curated Links & Resources — Collection of AI, CV, ML, and software development tools and tutorials — https://www.pirahansiah.com/notes/pkm/links/\n\n\nLink Collection\n\n\n \n \n Category\n Title\n Link\n Short Summary\n \n \n \n \n AI Agents\n Android Agent\n GitHub\n Open source Android AI agent.\n \n \n Audio\n SendBySound\n Website\n Data transfer through sound.\n \n \n GitHub\n bstnxbt\n GitHub\n Developer repositories.\n \n \n Python\n pyptx Docs\n Docs\n Installation guide for pyptx.\n \n \n AI Research\n Yann Dubs Repositories\n GitHub\n AI and ML projects.\n \n \n Computer Vision\n Ultralytics Hand Keypoints\n Docs\n Hand pose dataset documentation.\n \n \n AI Agents\n Tavily + OpenClaw\n Article\n AI agent tooling article.\n \n \n macOS\n CornerTasks\n GitHub\n Productivity utility for macOS.\n \n \n 3D\n MakeHuman Unity Export\n Website\n Exporting 3D models to Unity.\n \n \n Software Architecture\n Vertical Slice Architecture\n Article\n Software architecture pattern.\n \n \n AI Learning\n Andrej Karpathy Skills\n GitHub\n AI learning material collection.\n \n \n AI Agents\n Devin DeepWiki\n Docs\n Devin documentation.\n \n \n AI Agents\n Context Engineering Skills\n GitHub\n Agent skill framework.\n \n \n AI Agents\n Local Screen Agent\n GitHub\n Desktop AI agent ..." },{ "id": "/notes/docs/llm/llm-concepts/", "title": "Advanced LLM Concepts", "url": "/notes/docs/llm/llm-concepts/", "tags": ["ai","llm","rag","multimodal"], "hashtags": "#ai #llm #rag #multimodal", "category": "hub", "body": "\n Advanced LLM Concepts — Comprehensive mind map covering advanced LLM concepts including RAG, embeddings, multimodal models, and production deployment. — https://www.pirahansiah.com/notes/docs/llm/llm-concepts/\n\n\n\n\nMind Map: Orchestrating Agents & Advanced LLM Concepts\n\n1. Introduction\n\n Main Concept: Coordination of multiple AI agents for complex tasks, enhanced by Large Language Models (LLMs and multimodal technologies.\n Goals: Solve tasks beyond a single agent’s capability, using agents and LLMs in harmony.\n Key Technologies: LLMs, multimodal models (vision + text, task orchestration.\n\n\n2. Core Components of Agent Systems\n\n2.1 Agents\n\n Definition: Autonomous entities carrying out specific functions.\n Types:\n \n Single-purpose: Designed for specific tasks.\n General-purpose: Flexible agents capable of performing various tasks.\n \n \n Capabilities:\n \n Interaction with environments.\n Processing inputs and generating outputs.\n Self-contained decision-making.\n \n \n\n\n2.2 Orchestrator\n\n Definition: Central controller managing multiple agents and interacting with LLMs.\n Roles:\n \n Delegates tasks to agents.\n Monitors progress.\n Facilitates communication between agents and LLMs.\n Combines results for task completion.\n \n \n\n\n2.3 Communication Between Agents and LLMs\n\n Importance: Efficient communication enables effective agent-LLM collaboration.\n Methods:\n \n Message Passing between agents.\n API Calls to trigger LLMs for specific operations (like function calling.\n Shared Memory or datab..." },{ "id": "/notes/courses/metaverse/", "title": "Metaverse & XR", "url": "/notes/courses/metaverse/", "tags": ["metaverse","XR","spatial-computing","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "> **Metaverse & XR** — Workshop on extended reality, spatial computing, and immersive technologies. — https://www.pirahansiah.com/notes/courses/metaverse/\n\n# Metaverse & XR\n\nWorkshop on extended reality, spatial computing, and immersive technologies.\n\n#Metaverse #XR #SpatialComputing #FarshidPirahansiah\n" },{ "id": "/notes/courses/ml-spec/", "title": "Machine Learning Specialization", "url": "/notes/courses/ml-spec/", "tags": ["ML","Coursera","machine-learning-specialization","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "> **Machine Learning Specialization** — Coursera ML Specialization notes covering regression, classification, clustering, and neural networks. — https://www.pirahansiah.com/notes/courses/ml-spec/\n\n# Machine Learning Specialization 2022\n\nCoursera: Machine Learning Specialization (2022\n\nDownload source code and full text of Mind map: GitHub https://github.com/tiziran/tiziran.github.io/tree/main/Machine_Learning_Specialization\n\n## Course 1: Supervised Machine Learning: Regression and Classification\n\n- **Week 1:** Introduction to Machine Learning\n- **Week 2:** Regression with multiple input variables\n- **Week 3:** Classification\n\n## Course 2: Advanced Learning Algorithms\n\n- **Week 1:** Neural Networks\n- **Week 2:** Neural network training\n\n## Course 3: Unsupervised Learning, Recommenders, Reinforcement Learning\n\n- **Week 1-4:** Clustering, anomaly detection, recommender systems, reinforcement learning\n\n## Course 4: ML Clustering & Retrieval\n\n- Case Studies: Finding Similar Documents\n- K-nearest neighbors, KD-trees, locality sensitive hashing\n- K-means, expectation maximization (EM, latent Dirichlet allocation (LDA\n\n#Machine_Learning_Specialization #tiziran #FarshidPirahanSiah\n" },{ "id": "/notes/courses/mlops/", "title": "MLOps", "url": "/notes/courses/mlops/", "tags": ["MLOps","machine-learning-engineering","Coursera","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "> **MLOps** — Coursera MLOps Specialization covering ML production, data lifecycle, modeling pipelines, and deployment. — https://www.pirahansiah.com/notes/courses/mlops/\n\n# MLOps — Machine Learning Engineering for Production\n\nCoursera: Machine Learning Engineering for Production (MLOps Specialization\n\n## Course 1: Introduction to Machine Learning in Production\n- Week 1: Overview of the ML Lifecycle and Deployment\n- Week 2: Selecting and Training a Model\n- Week 3: Data Definition and Baseline\n\n## Course 2: Machine Learning Data Lifecycle in Production\n- Week 1: Collecting, Labeling, and Validating data\n- Week 2: Feature Engineering, Transformation, and Selection\n- Week 3: Data Journey and Data Storage\n- Week 4: Advanced Data Labeling Methods, Data Augmentation\n\n## Course 3: Machine Learning Modeling Pipelines in Production\n- Week 1: Neural Architecture Search (NAS, AutoML\n- Week 2: Model Resource Management (PCA, SVD, quantization, pruning\n- Week 3: High-Performance Modeling (distributed training, knowledge distillation\n- Week 4: Model Analysis (TFMA, TFX\n- Week 5: Interpretability (SHAP, LIME, PDP\n\n## Course 4: Deploying Machine Learning Models in Production\n- Week 1-2: Model Serving (TensorFlow Serving, TorchServe, Triton\n- Week 3: Model Management and Delivery\n- Week 4: Model Monitoring and Logging\n\n#MLOps #ComputerVision #Tiziran\n" },{ "id": "/notes/courses/modern-cpp/", "title": "Modern C++ for Image Processing", "url": "/notes/courses/modern-cpp/", "tags": ["C++","C++23","image-processing","design-patterns","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "> **Modern C++ for Image Processing** — Modern C++23 for image processing: memory management, design patterns, CUDA, and OpenCV. — https://www.pirahansiah.com/notes/courses/modern-cpp/\n\n# Advanced Programming with Modern C++ 23 for Image Processing\n\n## Key Topics\n\n### Memory Management\n- Call by value (stack vs call by reference (heap\n- Stack: LIFO, compile-time, 8MB limit, fast\n- Heap: manual management, runtime, dynamic, slower\n- Tools: valgrind, fsanitize\n\n### Design Patterns (Gang of Four\n**Creational (5:** Factory method, Abstract factory, Builder, Prototype, Singleton\n**Structural (7:** Adapter, Bridge, Composite, Decorator, Facade, Flyweight, Proxy\n**Behavioral (12:** Chain of responsibility, Command, Mediator, Observer, Interpreter, State, Strategy, Template method, Visitor, Iterator, Memento, Null-object\n\n### Smart Pointers\n- std::unique_ptr — exclusive ownership\n- std::shared_ptr — shared ownership\n- std::weak_ptr — non-owning reference\n- Rule of five for custom memory management\n\n### Compilation\n- CUDA Jetson: nvcc -std=c++14 -arch=sm_62 -o main.run main.cu\n- C++20 modules: clang++ -std=c++2a -c file.cpp -Xclang -emit-module-interface -o file.pcm\n\n## OpenCV C++ Tips\n- cv::Mat is a smart pointer — use .clone( for deep copy in vectors\n- imagesVector.push_back(imageMat.clone(;\n\n#OpenCV #C++ #tiziran #CPlusPlus\n" },{ "id": "/notes/docs/cv/multi-camera/", "title": "Real-Time Multi-Camera Vision Systems", "url": "/notes/docs/cv/multi-camera/", "tags": ["computer-vision","multi-camera","real-time","edge-ai"], "hashtags": "#cv #multicamera #realtime #edgeai", "category": "hub", "body": "> **Real-Time Multi-Camera Vision Systems** — Deep technical guide to building real-time multi-camera vision systems with heterogeneous processing and scaling to 100+ cameras. — https://www.pirahansiah.com/notes/docs/cv/multi-camera/\n\n# Real-Time Multi-Camera Vision Systems\n\nBuilding real-time multi-camera AI requires synchronizing 10–100 cameras while processing with CPU, GPU, NPU, and direct I/O in parallel. This guide covers frameworks (OpenCV, GStreamer, DeepStream, OpenVINO, low-level optimizations (USB DMA, UVC driver tweaks, and scaling strategies.\n\n---\n\n## 1. System Architecture\n\n### Pipeline\n\n1. **Capture** – RTSP, USB, MIPI-CSI input\n2. **Decode** – CPU, GPU hardware decoder, or FPGA\n3. **Preprocess** – resize, color convert, normalization\n4. **Inference** – DNNs on GPU/NPU/CPU\n5. **Post-process** – tracking, feature extraction\n6. **Output** – GUI, storage, or network stream\n\n### Framework Examples\n\n- **NVIDIA DeepStream**: GPU-accelerated multi-camera inference with batching (nvstreammux, inference (nvinfer, trackers, and OSD\n- **Intel OpenVINO**: Multi-Camera Multi-Target demo with detector + re-ID + tracker\n- **GStreamer**: Flexible pipelines with hardware decoders, multithreaded elements, and timestamp handling\n\n---\n\n## 2. Heterogeneous Processing (CPU/GPU/NPU\n\n Processor Role \n ----------- ------ \n CPU I/O, buffering, lightweight pre/post-processing \n GPU High-throughput DNN inference, CUDA/Vulkan processing \n NPU/DLA/TPU Dedicated ML accelerators for extra streams \n Multi-threading Double/triple buffering, async pipelines \n\nResearch sh..." },{ "id": "/notes/pubs/papers/multimodal-calibration/", "title": "Camera Calibration for Multi-Modal Vision", "url": "/notes/pubs/papers/multimodal-calibration/", "tags": ["camera-calibration","multi-modal","robotics","image-quality"], "hashtags": "#cameracalibration #multimodal #robotics #iqa", "category": "hub", "body": "> **Camera Calibration for Multi-Modal Vision** — Automatic calibration framework for multi-modal robot vision using IQA metrics. — https://www.pirahansiah.com/notes/pubs/papers/multimodal-calibration/\n\nCamera_Calibration_for_Multi-Modal_Robot_Vision\n\n\n\nhttps://www.pirahansiah.com/notes/pubs/papers/Camera_Calibration_for_Multi-Modal_Robot_Vision\n\n\nspotify https://podcasters.spotify.com/pod/show/pirahansiah/episodes/My-Conference-Paper-Camera-Calibration-for-Multi-Modal-Robot-Vision-e2ps1fn \n\nPDF Download My Conference Paper https://ieeexplore.ieee.org/document/7360336 \n\n\n{% if page.extname == \"camera-calibration-multi-modal.md\" %}\n !My Conference Paper Camera Calibration for Multi-Modal Robot Vision /notes/pubs/papers/camera-calibration-multi-modal.png\n{% else %}\n \n{% endif %}\n\n\n# Camera Calibration for Multi-Modal Robot Vision\n\n## 1. Introduction\n - **Objective**: Propose an automatic calibration method for multimodal robot vision.\n - **Challenges in Robot Vision**:\n - Image quality degradation\n - Difficulty in adjusting to different environments\n - **Key Issues**:\n - False negative data points due to poor calibration\n - Need for continuous recalibration in dynamic environments\n\n## 2. Image Quality and Calibration\n - **Image Quality Assessment (IQA**:\n - Impact of poor image quality on robot vision\n - Key metrics:\n - Peak Signal-to-Noise Ratio (PSNR\n - Structural Similarity Index (SSIM\n - **Camera Calibration (CC**:\n - Relationship between image quality and calibration accuracy\n - Need for automatic calibration t..." },{ "id": "/notes/pubs/books/ai/new-developer-era/", "title": "OpenCV 5 — Chapter 0: Introduction", "url": "/notes/pubs/books/ai/new-developer-era/", "tags": [], "hashtags": "", "category": "hub", "body": "> **OpenCV 5 — Chapter 0: Introduction** — Introduction to the new developer era, AI agents, and multi-agent systems. — https://www.pirahansiah.com/notes/pubs/books/ai/new-developer-era/\n\nreference index: index]]\ntags: ai-agents, multi-agent-systems, llm, deep-learning]\nhashtags: \"#aiagents #multiagent #llm #deeplearning\"\n\n{{ site.author }}\n\n# The New Developer Era: Transforming Your Career and Building Production Ready AI Agents in 2027\n\n- **Title Page** \n - *The New Developer Era: Transforming Your Career and Building Production Ready AI Agents in 2027; Agents will replace all software* \n - **Author Name**\n **Dr. Farshid Pirahansiah** \n Lead LLM/AI/Computer Vision Engineer & Researcher \n LinkedIn: linkedin.com/in/pirahansiah https://linkedin.com/in/pirahansiah\n\n - **Publication Details** \n - **Date of Publication**: 2027\n - **Edition**: Draft Edition \n\n - **Disclaimers**\n - **Informational Purposes**: The content of this book is based on the author’s professional experiences and research. It should not be taken as legal, financial, or specific career advice. Always perform due diligence before making major career or technical decisions. \n - **No Guarantees**: The author and publisher assume no responsibility for errors, omissions, or contrary interpretations of the subject matter. The views expressed are solely those of the author. \n - **Trademarks**: All product names, trademarks, and registered trademarks are property of their respective owners. \n - **Reproduction & Distribution**: No p..." },{ "id": "/notes/docs/cuda/numba/", "title": "Accelerate Python with Numba's @jit(nopython=True)", "url": "/notes/docs/cuda/numba/", "tags": ["python","numba","jit","performance"], "hashtags": "#python #numba #jit #performance", "category": "hub", "body": "> **Accelerate Python with Numba's @jit(nopython=True** — Tutorial on using Numba's @jit decorator to compile Python functions into optimized machine code for faster numerical computing. — https://www.pirahansiah.com/notes/docs/cuda/numba/\n\nAccelerate Python with Numba’s @jit(nopython=True \n\n\n\n# 🚀 Accelerate Python with Numba’s @jit(nopython=True 🚀\n\nAre you looking to **optimize your Python code** for better performance? If you work with **large datasets** or run complex numerical computations, the **Numba** library can be a game-changer!\n\nWith **@jit(nopython=True**, Numba translates Python functions into **machine code** using Just-In-Time (JIT compilation. This drastically reduces execution time, especially for **loops** and **numerical operations**.\n\nLet me show you how it works! 👇\n\n## 🚀 What is @jit(nopython=True?\n\n@jit(nopython=True is a decorator from the Numba library. It compiles the entire function into machine code at runtime. Here’s why it’s special:\n\n- **nopython=True**: Forces Numba to fully compile the function to machine code, skipping the Python interpreter. This ensures maximum performance.\n- It’s great for **numerical computing** or operations involving large arrays, matrices, or loops.\n\n💡 If Numba detects a dynamic type (like Python objects, it will throw an error with nopython=True, ensuring you stay in the compiled mode.\n\n---\n\n## 🛠️ Example: Summing Squares Without Numba\n\nLet’s start with a simple Python function that computes the sum of squares of a list of numbers.\n\npython\nimport time\n\n# Regular Python function (without Numba\ndef sum_of_squares(ar..." },{ "id": "/notes/pubs/books/ai/opencv5-ch1/", "title": "OpenCV 5 — Chapter 1: Image Basics", "url": "/notes/pubs/books/ai/opencv5-ch1/", "tags": [], "hashtags": "", "category": "hub", "body": "> **OpenCV 5 — Chapter 1: Image Basics** — Glossary of AI/ML terms, references, and resources for computer vision practitioners. — https://www.pirahansiah.com/notes/pubs/books/ai/opencv5-ch1/\n\ntags: opencv, image-processing, computer-vision, python]\nhashtags: \"#opencv #imageprocessing #cv #python\"\n\n{{ site.author }}\n# **Mastering AI and Machine Learning: Essential Glossary, Resources, and References**\n\nIn the rapidly evolving fields of Artificial Intelligence (AI and Machine Learning (ML, staying updated with key terms, resources, and the latest research is crucial for both novices and seasoned professionals. To support your journey, I’ve compiled a comprehensive guide encompassing a **Glossary of Key Terms**, **References and Further Reading**, and an **Index** of essential concepts and tools. Whether you're looking to deepen your understanding or seeking valuable resources to enhance your projects, this guide serves as a valuable roadmap.\n\n---\n\n## **Glossary of Key Terms**\n\nUnderstanding the foundational terminology is essential for navigating the complexities of AI and ML. Here are some pivotal terms:\n\n### **A**\n\n- **AI (Artificial Intelligence**: The simulation of human intelligence processes by machines, including learning, reasoning, and self-correction.\n- **Algorithm**: A step-by-step procedure or formula for solving a problem or accomplishing a task.\n- **API (Application Programming Interface**: A set of rules and protocols for building and interacting with software applications.\n\n### **B**\n\n- **Bias**: Systematic and unfair discrimination in AI model outcomes, o..." },{ "id": "/notes/pubs/books/ai/opencv5-ch2/", "title": "OpenCV 5 — Chapter 2: Feature Detection", "url": "/notes/pubs/books/ai/opencv5-ch2/", "tags": [], "hashtags": "", "category": "hub", "body": "> **OpenCV 5 — Chapter 2: Feature Detection** — Comprehensive guide to AI/ML concepts, tools, best practices, and ethical considerations. — https://www.pirahansiah.com/notes/pubs/books/ai/opencv5-ch2/\n\ntags: opencv, feature-detection, computer-vision, ai]\nhashtags: \"#opencv #featuredetection #cv #ai\"\n\n{{ site.author }}\n\n# **Mastering AI and Machine Learning: A Comprehensive Summary**\n\nArtificial Intelligence (AI and Machine Learning (ML have transformed the technological landscape, driving innovations across various industries. This summary encapsulates the essential concepts, tools, resources, and ethical considerations outlined in \"Mastering AI and Machine Learning,\" providing a foundational understanding for both novices and seasoned professionals.\n\n---\n\n## **1. Introduction to AI and Machine Learning**\n\n### **1.1 Understanding AI**\nAI refers to the simulation of human intelligence processes by machines, particularly computer systems. These processes include learning (the acquisition of information and rules for using the information, reasoning (using rules to reach approximate or definite conclusions, and self-correction.\n\n### **1.2 Machine Learning Fundamentals**\nML is a subset of AI focused on building systems that learn from data to improve their accuracy over time without being explicitly programmed. It encompasses various techniques, including supervised learning, unsupervised learning, and reinforcement learning.\n\n---\n\n## **2. Key Concepts and Terminology**\n\nA solid grasp of foundational terms is crucial for navigating AI and ML. Key concepts include:\n\n- **Neur..." },{ "id": "/notes/pubs/books/ai/opencv5-ch3/", "title": "OpenCV 5 — Chapter 3: Advanced Topics", "url": "/notes/pubs/books/ai/opencv5-ch3/", "tags": ["opencv","computer-vision","image-processing","filtering"], "hashtags": "#opencv #cv #imageprocessing #filtering", "category": "hub", "body": "> **OpenCV 5 — Chapter 3: Advanced Topics** — Advanced computer vision techniques including filtering, morphology, and G-API integration. — https://www.pirahansiah.com/notes/pubs/books/ai/opencv5-ch3/\n\n" },{ "id": "/notes/courses/openvino/", "title": "OpenVINO Deep Learning", "url": "/notes/courses/openvino/", "tags": ["OpenVINO","Intel","inference","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "> **OpenVINO Deep Learning** — Intel OpenVINO toolkit for optimizing and deploying deep learning models. — https://www.pirahansiah.com/notes/courses/openvino/\n\n# OpenVINO Deep Learning\n\nIntel OpenVINO toolkit for optimizing and deploying deep learning models on Intel hardware.\n\n#OpenVINO #Intel #Inference #FarshidPirahansiah\n" },{ "id": "/notes/docs/cv/optical-flow/", "title": "Optical Flow: Challenges and Solutions", "url": "/notes/docs/cv/optical-flow/", "tags": ["computer-vision","optical-flow","motion-analysis","opencv"], "hashtags": "#cv #opticalflow #motionanalysis #opencv", "category": "hub", "body": "> **Optical Flow: Challenges and Solutions** — Comprehensive guide to optical flow challenges including illumination changes, occlusions, fast motion, and their solutions. — https://www.pirahansiah.com/notes/docs/cv/optical-flow/\n\n1. Illumination Variations\n\n\t•\tProblem: Changes in lighting conditions can distort motion estimation.\n\t•\tSolution: Use robust optical flow algorithms like Lucas-Kanade with pyramids or deep learning models trained on diverse lighting conditions.\n \n\n\t•\tChallenge: Changes in lighting distort motion estimation.\n\t•\tSolution:\n\t•\tUse algorithms robust to illumination changes, such as:\n\t•\tHorn-Schunck with brightness constancy assumption modifications.\n\t•\tAdvanced methods like FlowNet2 or RAFT trained on diverse lighting conditions.\n\t•\tNormalize pixel intensities (e.g., histogram equalization or use illumination-invariant feature descriptors.\n\t•\tFunction/Algorithm Examples:\n\t•\tNormalized Cross-Correlation (NCC for robust feature matching under varying lighting.\n\t•\tIllumination-Invariant Optical Flow in the Horn-Schunck model (custom implementations exist.\n\t•\tDeep learning models like RAFT and PWC-Net, trained on diverse lighting.\n\n\t•\tOpenCV functions:\n\t•\tcv2.calcOpticalFlowFarneback(: Dense optical flow with Gaussian filtering (robust under moderate illumination changes.\n\t•\tcv2.createCLAHE(: Apply Contrast Limited Adaptive Histogram Equalization for preprocessing to normalize illumination.\n\t•\tcv2.equalizeHist(: Normalize brightness and contrast in grayscale images.\n\n\n\n\n Illumination Variations in Motion Estimation\nSummary\nRobust motion estimation t..." },{ "id": "/notes/courses/parallel/", "title": "Parallel Programming for Computer Vision", "url": "/notes/courses/parallel/", "tags": ["parallel-programming","Python","computer-vision","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "> **Parallel Programming for Computer Vision** — Python parallel programming techniques for computer vision applications. — https://www.pirahansiah.com/notes/courses/parallel/\n\n# Parallel Programming\n\nPython Parallel Programming for computer vision applications.\n\nTopics covered:\n- Multi-threading and multi-processing\n- GPU acceleration with CUDA\n- Distributed computing\n- Performance optimization for CV pipelines\n\n#computervision #parallelprogramming #FarshidPirahansiah\n" },{ "id": "/notes/pubs/papers/pattern-calibration/", "title": "Pattern Image for Camera Calibration", "url": "/notes/pubs/papers/pattern-calibration/", "tags": ["camera-calibration","pattern-recognition","robotics"], "hashtags": "#cameracalibration #patternrecognition #robotics", "category": "hub", "body": "> **Pattern Image for Camera Calibration** — Analysis of how pattern image design impacts camera calibration accuracy and reliability. — https://www.pirahansiah.com/notes/pubs/papers/pattern-calibration/\n\nPattern_Image_Significance_for_Camera_Calibration\n\n\nhttps://www.pirahansiah.com/notes/pubs/papers/Pattern_Image_Significance_for_Camera_Calibration\n\n\nspotify https://podcasters.spotify.com/pod/show/pirahansiah/episodes/Pattern-Image-Significance-for-Camera-Calibration-e2ps2mt \n\nPDF Download My Conference Paper http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8305440&isnumber=8305342 \n\n\n{% if page.extname == \"pattern-image-calibration.md\" %}\n !My Conference Paper Pattern Image Significance for Camera Calibration /notes/pubs/papers/pattern-image-calibration.png\n{% else %}\n \n{% endif %}\n\n\n# Pattern Image Significance for Camera Calibration\n\n## 1. Introduction\n - **Objective**: Discuss the significance of pattern images in camera calibration.\n - **Camera Calibration**:\n - A method to estimate the parameters of a pinhole camera model.\n - Used to correct lens distortion and for 3D reconstruction in applications like machine vision, robotics, and navigation systems.\n\n## 2. Categories of Camera Calibration Methods\n - **Self-Calibration**:\n - Does not rely on a known calibration object.\n - Nonlinear, sensitive to noise, requires more computational power.\n - **Active Vision Calibration**:\n - Based on controlled camera motion.\n - Solves parameters by using images captured during known motions.\n - **Known Object Calibration**:\n - Tr..." },{ "id": "/notes/slides/presentation/", "title": "Reducing Token Usage in AI-Assisted Development", "url": "/notes/slides/presentation/", "tags": ["presentation","ai","llm","token-optimization"], "hashtags": "#presentation #ai #llm #token-optimization", "category": "hub", "body": "> **Reducing Token Usage in AI-Assisted Development** — Reducing Token Usage in AI-Assisted Development — https://www.pirahansiah.com/notes/slides/presentation/\n\n\n\n\n\n\n\n ← tap back tap next →\n \n \n\n \n Reducing Token Usage\n in AI-Assisted Development\n Dr. Farshid Pirahansiah\n \n\n \n The Problem\n \n 162 KBFull .claude folder\n 100%Token cost\n SlowResponse time\n $$$API cost\n \n 1 token ≈ 4 chars • Every file = tokens\n \n\n \n Token Cost by Component\n \n ~10 KBCore files✅ Always needed\n ~90 KBSkills (32 filesSelect per project\n ~14 KBAgents (7 filesSelect per domain\n 6-10%Minimal configBest savings\n \n \n\n \n Strategy 1: .cursorignore\n \n 5-10%MinimalCore only\n 20-30%WebPython/backend\n 35-45%CV/MLYOLO, SAM2\n 40-50%Edge/C++Inference\n \n Copy template → project/.cursorignore\n \n\n \n Strategy 2: Selective Loading\n \n \n ✅ INCLUDE\n ! .claude/skills/cv-pipeline/\n ! .claude/agents/debugger.md\n ! .claude/CLAUDE.md\n \n \n ❌ EXCLUDE\n .claude/skills/portfolio/\n .claude/workflows/\n .claude/agent-memory/\n \n \n 38-45% instead of 100%\n \n\n \n Strategy 3: Remove Files\n \n \n Delete\n DIRECTORY-TREE.md\n FILE-INVENTORY.md\n PROJECT_PORTFOLIO ..." },{ "id": "/privacy/", "title": "Privacy Policy", "url": "/privacy/", "tags": [], "hashtags": "", "category": "hub", "body": "# Privacy Policy\n\n**Effective date:** 2025-11-30\n\n## 1. Who We Are\n\nThis website is operated by Dr. Farshid Pirahansiah.\n\n## 2. What Data We Collect\n\n- IP address and browser info (via analytics\n- Device and usage data (via cookies and Google Ads\n- Any data you enter in forms (e.g., contact forms, sign-ups\n\n## 3. Use of Google Services\n\nWe use Google services to deliver and measure content and advertising on this site. Specifically:\n\n- **Google AdSense / Ads:** We display ads using Google AdSense. Our AdSense Publisher ID is pub-7548808026663917. Google may use cookies and other identifiers to show personalized ads and to measure ad performance. For details see Google's advertising technologies https://policies.google.com/technologies/ads.\n- **Google Analytics:** We use Google Analytics to collect aggregated information about how visitors use the site. This helps us improve the website and content.\n\n## 4. Cookie Policy\n\nWe use cookies to:\n\n- Remember your preferences\n- Analyze site traffic\n- Serve personalized advertisements through Google AdSense\n\nYou can control cookies through your browser settings.\n\n## 5. Your Rights (GDPR\n\nIf you are located in the European Economic Area (EEA, you have the following rights under GDPR:\n- **Right of access** — Request a copy of the data we hold about you.\n- **Right to rectification** — Request correction of inaccurate data.\n- **Right to erasure** — Request deletion of your personal data.\n- **Right to restrict processing** — Request limitation of how we use your data.\n- **Right to data portability** — Receive your data in a structured,..." },{ "id": "/notes/docs/product/", "title": "Product — Embedded CV & Edge AI", "url": "/notes/docs/product/", "tags": ["computer-vision","edge-ai","embedded-systems","product"], "hashtags": "#cv #edgeai #embeddedsystems #product", "category": "hub", "body": "> **Product — Embedded CV & Edge AI** — Embedded computer vision and edge AI systems portfolio. — https://www.pirahansiah.com/notes/docs/product/\n\n\n\n\n Product\n Embedded computer vision & edge AI systems\n\n\n\n ← Home\n Research\n Solutions\n\n\n\n\n \n 👁\n Computer Vision Core\n \n → 3D Vision & Multi-Camera\n → Optical Flow\n → Multi-Camera Systems\n → CV Coaching Roadmap\n \n \n\n \n 🤖\n AI & LLMs New\n \n → Advanced LLM Concepts\n → Orchestrating AI Agents\n → AI Blog\n → Avatar Generator\n \n \n\n \n ⚡\n CUDA & GPU Dev\n \n → Numba JIT Tutorial\n → PyCUDA Kernels\n → CUDA in VS Code\n → MLX, CoreML & Metal\n \n \n\n \n ⚙\n Optimization ML\n \n → CV/DL/ML Optimization\n → Prompt Engineering\n \n \n\n \n 💻\n Programming\n \n → C++ Quick Reference\n → Python Configuration\n → Developer Tools\n → Shell & Vim\n \n \n\n\n" },{ "id": "/notes/pkm/proof/", "title": "Site Link Index", "url": "/notes/pkm/proof/", "tags": ["site-index","links","documentation"], "hashtags": "#siteindex #links #documentation", "category": "pkm", "body": "> **Site Link Index** — Complete index of all pages, publications, and resources on pirahansiah.com — https://www.pirahansiah.com/notes/pkm/proof/\n\n# Pirahansiah.com — Complete Link Index\n\nAll content files are under notes/. Links below reference local files only.\n\n---\n\n## Landing & Profile\n\n File Path Description \n ------ ------ ------------- \n Home /notes/docs/ /notes/docs/ Main landing page \n Coaching /notes/docs/coaching/ /notes/docs/coaching/ CV coaching and teaching roadmap \n Resources /notes/docs/resources/ /notes/docs/resources/ Patents, books, papers, projects \n Links /notes/docs/links/ /notes/docs/links/ Curated links and tools \n\n---\n\n## Computer Vision & 3D\n\n File Path Description \n ------ ------ ------------- \n 3D Vision /notes/docs/cv/3d/ /notes/docs/cv/3d/ Point clouds, depth sensing, multi-camera sync \n Optical Flow /notes/docs/cv/optical-flow/ /notes/docs/cv/optical-flow/ Motion estimation challenges and solutions \n Multi-Camera /notes/docs/cv/multi-camera/ /notes/docs/cv/multi-camera/ Scaling to 100+ cameras \n\n---\n\n## AI & LLMs\n\n File Path Description \n ------ ------ ------------- \n Advanced LLM Concepts /notes/docs/llm/llm-concepts/ /notes/docs/llm/llm-concepts/ RAG, embeddings, multimodal \n Orchestrating Agents /notes/docs/llm/agents/ /notes/docs/llm/agents/ Multi-agent systems \n Blog /notes/docs/llm/blog/ /notes/docs/llm/blog/ RAG vs CAG, multi-agent architectures \n Avatar Generator /notes/docs/llm/avatar/ /notes/docs/llm/avatar/ Ollama + Wav2Lip tutorial \n\n---\n\n## CUD..." },{ "id": "/notes/pubs/journals/psnr-segmentation/", "title": "PSNR Threshold for Image Segmentation", "url": "/notes/pubs/journals/psnr-segmentation/", "tags": ["psnr","thresholding","image-segmentation","image-processing"], "hashtags": "#psnr #thresholding #imagesegmentation #imageprocessing", "category": "hub", "body": "> **PSNR Threshold for Image Segmentation** — PSNR-based thresholding method for image segmentation in automated visual inspection. — https://www.pirahansiah.com/notes/pubs/journals/psnr-segmentation/\n\nPEAK_SIGNAL-TO-NOISE_RATIO_BASED_ON_THRESHOLD_METHOD_FOR_IMAGE_SEGMENTATION\n\nhttps://www.pirahansiah.com/notes/pubs/journals/PEAK_SIGNAL-TO-NOISE_RATIO_BASED_ON_THRESHOLD_METHOD_FOR_IMAGE_SEGMENTATION/\n\nspotify https://podcasters.spotify.com/pod/show/pirahansiah/episodes/PEAK-SIGNAL-TO-NOISE-RATIO-BASED-ON-THRESHOLD-METHOD-FOR-IMAGE-SEGMENTATION-e2prep9\n\nPDF Download http://www.jatit.org/volumes/Vol57No2/4Vol57No2.pdf \n\n\n{% if page.extname == \"psnr-threshold-segmentation.md\" %}\n !PEAK SIGNAL-TO-NOISE RATIO BASED ON THRESHOLD METHOD FOR IMAGE SEGMENTATION /notes/pubs/journals/psnr-threshold-segmentation.png\n{% else %}\n \n{% endif %}\n\n\n# Mind Map: Peak Signal-to-Noise Ratio Based on Threshold Method for Image Segmentation\n\n## 1. Introduction\n- **Importance of Thresholding**: Separates objects from the background, crucial in image processing and pattern recognition.\n- **Types of Thresholding**:\n - **Single Thresholding**: Produces binary images (0 and 1.\n - **Multilevel Thresholding**: Produces images with pixel values between 0 and 255.\n- **Objective**: Develop a new algorithm using Peak Signal-to-Noise Ratio (PSNR for image segmentation.\n\n## 2. Thresholding Techniques\n### 2.1 Single Thresholding\n- **Definition**: Uses a single threshold value to convert the image into binary.\n- **Goal**: Maximize segmentation accuracy and reduce storage requirements.\n- **Methods**:\n - ..." },{ "id": "/notes/docs/cuda/pycuda/", "title": "How PyCUDA Reads and Runs C Kernels", "url": "/notes/docs/cuda/pycuda/", "tags": ["cuda","gpu","python","parallel-computing"], "hashtags": "#cuda #gpu #python #parallelcomputing", "category": "hub", "body": "> **How PyCUDA Reads and Runs C Kernels** — Explanation of how PyCUDA compiles and executes CUDA kernels written in C from Python, with step-by-step examples. — https://www.pirahansiah.com/notes/docs/cuda/pycuda/\n\nHow PyCUDA Reads and Runs C Kernels\n\n\n\n# How PyCUDA Reads and Runs C Kernels\n\nIn **PyCUDA**, you can run **CUDA kernels** (which are typically written in C or C++ directly from Python. PyCUDA provides a way to write CUDA code as a string, compile it at runtime, and execute it on the GPU. Let's walk through the process step-by-step, explaining how PyCUDA interacts with a kernel written in C and runs it.\n\n## 1. Importing PyCUDA Modules\nThe following lines import PyCUDA’s functionalities:\n\npython\nimport pycuda.curandom as curandom\nimport pycuda.driver as cuda\nimport pycuda.autoinit\n\n\n- **pycuda.driver as cuda**: This module provides the basic interface to communicate with the CUDA driver, which manages GPU resources and executes code.\n- **pycuda.autoinit**: This module automatically initializes CUDA when you import it, setting up the GPU and its context (the memory space for the program to run.\n- **pycuda.curandom**: This module is used to generate random numbers on the GPU using CUDA’s random number generation capabilities.\n\n## 2. Importing the CUDA Kernels\n\npython\nimport GraphTsetlinMachine.kernels as kernels\n\n\n- This imports a module named **kernels** from GraphTsetlinMachine. Presumably, this module contains CUDA code (written in C in the form of strings or functions that will later be compiled and executed using PyCUDA.\n\n## 3. Writing CUDA Kernels (C Code in ..." },{ "id": "/notes/docs/research/", "title": "Research — Publications & Patents", "url": "/notes/docs/research/", "tags": ["research","publications","patents","computer-vision"], "hashtags": "#research #publications #patents #cv", "category": "hub", "body": "> **Research — Publications & Patents** — Complete list of patents, journal articles, conference papers, books, and keynotes. — https://www.pirahansiah.com/notes/docs/research/\n\n\n\n\n Research\n 3 patents, 2 books, 6 journals, 11 conference papers, 1 keynote\n\n\n\n ← Home\n Product\n Solutions\n\n\n\n\n \n 📜\n Patents 3\n \n → Face Image Augmentation — WO 2021/060971\n → Vehicle Detection — WO 2021/107761\n → Facial Analysis Advertising — WO 2020/141969\n \n \n\n \n 📅\n Conference Papers 11\n \n → Adaptive Segmentation PSNR\n → License Plate — Entropy\n → Multi-threshold License Plate\n → Thresholding Handwritten\n → Camera Calibration Multi-Modal\n → Pattern Image Calibration\n → 2D vs 3D Map Movement\n → Character Recognition\n → Classification Geometrical\n → TafreshGrid\n → All Papers →\n \n \n\n \n 📄\n Journal Articles 6\n \n → Adaptive Thresholding PSNR\n → GSFT-PSNR Fuzzy Threshold\n → PSNR Threshold Segmentation\n → Character & Object Recognition\n → 3D SLAM & Humanoid Robots\n → Ant Colony Optimization\n \n \n\n \n 📖\n Books & Chapters 7\n \n → Computer Vision Meets LLM\n → OpenCV 5 — Ch.1 Image Basics\n → OpenCV 5 — Ch.2 Feature Detection\n → OpenCV 5 — Ch.3 Advanced Topics\n → Optical Flow for Video Stabilization\n → Camera Calibration ..." },{ "id": "/notes/docs/resources/", "title": "Portfolio & Publications", "url": "/notes/docs/resources/", "tags": ["portfolio","publications","patents","research"], "hashtags": "#portfolio #publications #patents #research", "category": "hub", "body": "> **Portfolio & Publications** — Complete portfolio of patents, publications, books, and project highlights. — https://www.pirahansiah.com/notes/docs/resources/\n\n# My Portfolio\n\n- Impact Portfolio /notes/docs/projects/solutions/\n\n# Resume\n\n- Dr. Farshid Pirahansiah CV /notes/pubs/cv/\n\n## My Publications\n\n### My Patents (3\n\n- A METHOD FOR AUGMENTING A PLURALITY OF FACE IMAGES WO2021060971A1 /notes/pubs/patents/face-augmentation/\n- SYSTEM AND METHOD FOR PROVIDING ADVERTISEMENT CONTENTS BASED ON FACIAL ANALYSIS WO2020141969A2 /notes/pubs/patents/facial-analysis/\n- A METHOD FOR DETECTING A MOVING VEHICLE WO2021107761A1 /notes/pubs/patents/vehicle-detection/\n\n### My Books (2\n\n- Computational Intelligence: From Theory to Application /notes/pubs/books/computational-intelligence/\n- Camera Calibration and Video Stabilization Framework for Robot Localization — Springer /notes/pubs/books/camera-calibration/\n\n### My Journals (6\n\n- Adaptive Image Thresholding Based on PSNR /notes/pubs/journals/adaptive-thresholding/\n- CHARACTER AND OBJECT RECOGNITION BASED ON GLOBAL FEATURE EXTRACTION /notes/pubs/journals/character-recognition/\n- GSFT-PSNR Global Single Fuzzy Threshold /notes/pubs/journals/fuzzy-thresholding/\n- PEAK SIGNAL-TO-NOISE RATIO BASED ON THRESHOLD METHOD FOR IMAGE SEGMENTATION /notes/pubs/journals/psnr-segmentation/\n- 3D SLAM Simultaneous Localization And Mapping /notes/pubs/journals/slam-humanoid/\n- USING AN ANT COLONY OPTIMIZATION ALGORITHM /notes/pubs/journals/ant-colony/\n\n### My Conference Papers (11\n\n- 2D versus 3D Map for Environment Movement Objects /notes/pubs/papers..." },{ "id": "/notes/courses/risc-v/", "title": "RISC-V for AI", "url": "/notes/courses/risc-v/", "tags": ["RISC-V","edge-AI","processor","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "> **RISC-V for AI** — RISC-V open-source processor architecture for AI at the edge. — https://www.pirahansiah.com/notes/courses/risc-v/\n\n# RISC-V for AI\n\n## Key References\n- RISC-V Magazine (December 2020\n- RISC-V Summit 2020\n\n## Resources\n- riscv.org https://riscv.org/\n- HiFive1 Rev B https://www.amazon.de/-/en/HiFive1-Rev-B/dp/B086RGFS5N/\n- RISC-V 15 minute sample course https://www.eenewseurope.com/news/risc-v-boom-edge-ai-says-facebooks-chief-ai-scientist\n\n#RISC-V #EdgeAI #FarshidPirahansiah\n" },{ "id": "/notes/courses/ros/", "title": "ROS: Robot Operating System", "url": "/notes/courses/ros/", "tags": ["ROS","robotics","ETH-Zurich","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "> **ROS: Robot Operating System** — ETH Zurich lecture on ROS architecture, navigation, simulation, and robot models. — https://www.pirahansiah.com/notes/courses/ros/\n\n# ROS — Robot Operating System\n\nSource: ETH Zurich Lecture https://www.youtube.com/watch?v=0BxVPCInS3M\n\nLecturers: Péter Fankhauser, Dominic Jud, Martin Wermelinger\n\n## Topics\n- ROS Architecture & Navigation\n- ROS System & Packages\n- Simulating ROS with Gazebo\n- Robot Models (URDF, SDF\n- Simulation Environments\n- Visualization (RViz, rqt\n\n#ROS #Robotics #Tiziran\n" },{ "id": "/notes/docs/shell-vim/", "title": "Shell & Vim Quick Reference", "url": "/notes/docs/shell-vim/", "tags": ["shell","vim","cli","developer-tools"], "hashtags": "#shell #vim #cli #devtools", "category": "hub", "body": "> **Shell & Vim Quick Reference** — Quick reference for shell commands, vim basics, and essential CLI tools. — https://www.pirahansiah.com/notes/docs/shell-vim/\n\n- NeoHtop https://github.com/Abdenasser/neohtop\n- \n\n\n- The Shell\n - echo\n - \"\" \\ \n - $PATH\n - cd -\n \n - ctrl+L to clean \n - ctrl+r\n\n - >> append\n - output into input\n - # into root or sudo su\n - xdg-open opent the file with reletive app\n - foo=bar ; must be without space \n - echo \"value is $foo\" -> value is bar\n - echo 'value is $foo' -> value is $foo\n - mcd ( {\n mkdir -p \"$1\"\n cv \"$1\"\n }\n source mcd.sh\n !!\n\ncommand + /\noption+shif+a\n\n\n\n\n#!/usr/bin/env python\nimport sys\nfor arg in reversed(sys.argv1:]:\n print(arg\n\n# tools\ntldr\nlocate \ngrep \nripgrep = rg\nfzf\nbroot\nnnn\n\n\n# vim\n i\n esc\n r\n k\n s-v\n c-v\n :\n" },{ "id": "/notes/sitemap/", "title": "Sitemap — All Pages", "url": "/notes/sitemap/", "tags": [], "hashtags": "", "category": "hub", "body": "\n\n\n Sitemap\n Navigate through all pages on pirahansiah.com\n\n\n\n 🔍\n \n\n\n\n {{ site.pages size }}Pages\n 3Patents\n 6Journals\n 11Papers\n\n\n\n\n \n \n 🏠\n Home Landing\n \n → Dr. Farshid Pirahansiah\n \n \n\n \n \n 👁\n Computer Vision AI\n \n → 3D Vision & Multi-Camera\n → Optical Flow\n → Multi-Camera Systems\n → CV Coaching Roadmap\n \n \n\n \n \n 🤖\n AI & LLMs New\n \n → Advanced LLM Concepts\n → Orchestrating AI Agents\n → Blog: AI & LLMs\n → Avatar Generator\n \n \n\n \n \n ⚡\n CUDA & GPU Dev\n \n → Numba JIT Tutorial\n → PyCUDA Kernels\n → CUDA in VS Code\n → MLX, CoreML & Metal\n \n \n\n \n \n 💻\n Programming Tools\n \n → C++ Quick Reference\n → Python Configuration\n → Developer Tools & Setup\n → Shell & Vim Reference\n \n \n\n \n \n ⚡\n Optimization ML\n \n → CV, DL & ML Optimization\n → Prompt Engineering\n \n \n\n \n \n 🚀\n Business & Career\n \n → Startup Guide\n → SEO for LLMs\n → Top LinkedIn Posts 2024\n \n \n\n \n \n 📚\n Resources\n \n → Portfolio & Publications\n → Curated Links\n → Content Hub\n \n \n\n \n \n 📜\n Patents 3\n \n → All Patents\n → Face Image Augmentation\n → Vehicle Detection\n → Facial Analysis Advertising\n \n \n\n ..." },{ "id": "/notes/pubs/journals/slam-humanoid/", "title": "3D SLAM and Humanoid Robot Linkages", "url": "/notes/pubs/journals/slam-humanoid/", "tags": ["slam","humanoid-robots","3d-mapping","robotics"], "hashtags": "#slam #humanoidrobots #3dmapping #robotics", "category": "hub", "body": "> **3D SLAM and Humanoid Robot Linkages** — Survey of 3D SLAM techniques for localization, mapping, and humanoid robot applications. — https://www.pirahansiah.com/notes/pubs/journals/slam-humanoid/\n\n3D_SLAM_Simultaneous_Localization_And_Mapping_Trends_And_Humanoid_Robot_Linkages\n\nhttps://www.pirahansiah.com/notes/pubs/journals/3D_SLAM_Simultaneous_Localization_And_Mapping_Trends_And_Humanoid_Robot_Linkages/\n\nspotify https://podcasters.spotify.com/pod/show/pirahansiah/episodes/3D-SLAM-Simultaneous-Localization-And-Mapping-Trends-And-Humanoid-Robot-Linkages-e2prg07\n\nPDF Download http://journalarticle.ukm.my/6644/1/4429-10302-1-SM.pdf \n\n\n{% if page.extname == \"3d-slam-humanoid-robots.md\" %}\n !3D SLAM Simultaneous Localization And Mapping Trends And Humanoid Robot Linkages /notes/pubs/journals/3d-slam-humanoid-robots.png\n{% else %}\n \n{% endif %}\n\n\n# Simultaneous Localization and Mapping Trends and Humanoid Robot Linkages\n\n## 1. Introduction\n- SLAM: Simultaneous Localization and Mapping\n - Real-time map creation and localization\n - Robotics application: goal determination, motion planning\n - Usage in rescue missions, medical field, pipeline inspection, and more\n - Challenges: sensor uncertainty, correspondence, loop closing, time complexity\n\n## 2. SLAM Methods\n- Kalman Filter (KF\n - Bayesian filter handling uncertainty\n - Extended KF (EKF, Unscented KF (UKF, SEIF improvements\n - Challenges: computational resources, landmark growth\n- Particle Filter\n - Non-parametric recursive algorithm\n - Handles non-linearity and non-Gaussian noise\n - FastSLAM algorithms: O(P lo..." },{ "id": "/notes/docs/solutions/", "title": "Solutions — AI Courses & Workshops", "url": "/notes/docs/solutions/", "tags": ["ai","courses","workshops","deep-learning"], "hashtags": "#ai #courses #workshops #deeplearning", "category": "hub", "body": "> **Solutions — AI Courses & Workshops** — AI 2026 courses, workshops, and business solutions for production-ready products. — https://www.pirahansiah.com/notes/docs/solutions/\n\n\n\n\n Solutions\n Turn AI research into production-ready products\n\n\n\n ← Home\n Product\n Research\n\n\n\n\n \n 🚀\n AI 2026 — Courses 9\n \n → ML Specialization\n → Full Stack Deep Learning\n → FSDL 2022\n → MLOps\n → ROS\n → Parallel Programming\n → Modern C++\n → Cloud-Native\n → TensorFlow Deployment\n \n \n\n \n 📚\n Workshops 7\n \n → RISC-V\n → Edge AI Summit\n → Embedded IoT\n → Tesla AI\n → AI Hardware\n → OpenVINO\n → Metaverse\n \n \n\n \n 💡\n Business Growth\n \n → Startup Guide\n → SEO for LLMs\n → Prompt Engineering\n \n \n\n \n 📚\n Resources All\n \n → Curated Links\n → Top LinkedIn Posts\n → Book Summaries\n → IoT Scholarship\n → All Topics\n \n \n\n \n 📝\n Connect\n \n → LinkedIn\n → GitHub (43 repos\n → Full Sitemap\n \n \n\n\n" },{ "id": "/notes/docs/startup/", "title": "Startup Guide: Edge AI Business", "url": "/notes/docs/startup/", "tags": ["startup","business","fundraising","ai"], "hashtags": "#startup #business #fundraising #ai", "category": "hub", "body": "> **Startup Guide: Edge AI Business** — Edge AI business plan, fundraising in Germany, pitch decks, and growth strategies. — https://www.pirahansiah.com/notes/docs/startup/\n\n# Edge AI Solutions: Local LLM Implementation Business Plan\n\n## Executive Summary\n\nEdge AI Solutions will provide fully local, on-premises large language model (LLM deployments for businesses concerned with privacy, data security, compliance, and cost efficiency. By deploying powerful AI models directly on customer hardware, we eliminate cloud dependency, API costs, and data privacy concerns while delivering high-performance AI capabilities.\n\nKey value propositions:\n- Complete data privacy with no external connections\n- No recurring token costs or API fees\n- Hardware solutions ranging from $5,000 to $500,000\n- Customizable AI capabilities for specific industry needs\n- Compliance with strict data protection regulations\n\nOur target market includes industries with sensitive data requirements (healthcare, legal, finance, government and businesses seeking cost-effective AI implementation without ongoing usage fees.\n\n## Business Model Overview\n\n### Product Offerings\n\n1. **Edge AI Hardware Packages**:\n - Entry-level solutions: $5,000-20,000\n - Mid-tier enterprise solutions: $20,000-100,000\n - High-performance data center solutions: $100,000-500,000\n\n2. **Setup and Deployment Services**:\n - Professional installation: $1,000 per system\n - Custom model fine-tuning: $5,000-25,000 depending on requirements\n - Network integration: $2,000-5,000\n\n3. **Rental/Subscription Option**:\n - Monthly rental at..." },{ "id": "/terms/", "title": "Terms of Service", "url": "/terms/", "tags": [], "hashtags": "", "category": "hub", "body": "> **Terms of Service** — Terms of Service — https://www.pirahansiah.com/terms/\n\n# Terms of Service\n\n**Effective date:** 2025-11-30\n\n## 1. Acceptance of Terms\n\nBy accessing this website (pirahansiah.com, you agree to be bound by these Terms of Service. If you do not agree with any part of these terms, please do not use this website.\n\n## 2. Content\n\nAll content on this site — including articles, research summaries, code examples, course notes, and publications — is provided for informational and educational purposes only. Dr. Farshid Pirahansiah makes reasonable efforts to ensure accuracy but provides no warranties regarding the completeness, reliability, or suitability of the content.\n\n## 3. Intellectual Property\n\nAll content, code, designs, and materials on this site are protected by copyright law. You may:\n- Read and share links to pages\n- Quote brief excerpts with attribution\n\nYou may not:\n- Reproduce, distribute, or republish full articles without written permission\n- Use content for commercial purposes without authorization\n- Remove copyright notices from any materials\n\n## 4. Third-Party Services\n\nThis website uses third-party services including:\n- **Google AdSense** for displaying advertisements\n- **Google Analytics** for traffic analysis\n- **GitHub Pages** for hosting\n\nThese services may collect data according to their own privacy policies. See our Privacy Policy /privacy/ for details on how we handle your data.\n\n## 5. Links to External Sites\n\nThis website contains links to external sites (academic papers, GitHub repositories, LinkedIn, etc.. We are not responsible..." },{ "id": "/notes/courses/tesla/", "title": "Tesla AI", "url": "/notes/courses/tesla/", "tags": ["Tesla","autonomous-driving","AI","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "> **Tesla AI** — Workshop on Tesla's autonomous driving technology and AI systems. — https://www.pirahansiah.com/notes/courses/tesla/\n\n# Tesla AI\n\nWorkshop on Tesla's autonomous driving technology and AI systems.\n\n#Tesla #AutonomousDriving #FarshidPirahansiah\n" },{ "id": "/notes/courses/tf-deploy/", "title": "TensorFlow Deployment Specialization", "url": "/notes/courses/tf-deploy/", "tags": ["TensorFlow","TF.js","TF-Lite","mobile-ML","FarshidPirahansiah"], "hashtags": "", "category": "hub", "body": "> **TensorFlow Deployment Specialization** — TensorFlow deployment: browser-based TF.js, device-based TF Lite, data pipelines, and optimization. — https://www.pirahansiah.com/notes/courses/tf-deploy/\n\n# TensorFlow: Data and Deployment Specialization\n\n## 4 Courses\n\n1. **Browser-based Models with TensorFlow.js**\n2. **Device-based Models with TensorFlow Lite**\n3. **Data Pipelines with TensorFlow Data Services**\n4. **Advanced Deployment Scenarios with TensorFlow**\n\n## TF Lite Benefits\n- Lightweight + low-latency + privacy\n- Improved power consumption\n- Efficient model ready to use\n\n## Optimization Techniques\n- Quantization (all CPU platforms, reduced latency, low memory\n- Weight pruning\n- Model topology transforms\n- Tensor decomposition\n- Distillation\n\n#TensorFlow #EdgeAI #MobileML #FarshidPirahansiah\n" },{ "id": "/notes/pkm/use-cases/", "title": "Complete Portfolio", "url": "/notes/pkm/use-cases/", "tags": ["portfolio","computer-vision","edge-ai","github-projects"], "hashtags": "#portfolio #computervision #edgeai #github", "category": "pkm", "body": "> **Complete Portfolio** — Publications, patents, expertise, tools, and GitHub projects across CV and AI — https://www.pirahansiah.com/notes/pkm/use-cases/\n\n# Table of Contents\n\n- About #about\n- Publications #publications\n - Patents #patents\n - Books #books\n - Journals #journals\n - Conference Papers #conference-papers\n - Keynotes #keynotes\n- Camera Calibration Expertise #camera-calibration-expertise\n- Hardware & Platforms #hardware--platforms\n- Products & Tools #products--tools\n- Open Source #open-source\n- Technical Content #technical-content\n - CUDA & GPU Programming #cuda--gpu-programming\n - Optical Flow #optical-flow\n - 3D Vision & Multi-Camera #3d-vision--multi-camera\n - Optimization #optimization\n - AI & LLM #ai--llm\n- CV Coaching Roadmap #cv-coaching-roadmap\n- Courses #courses\n- Workshops & Events #workshops--events\n- GitHub Projects Portfolio #github-projects-portfolio\n\n---\n\n# About\n\nDr. Farshid Pirahansiah — Computer Vision expert with 12+ years R&D. 3 AI patents, 141+ citations (h-index 7, Springer book chapter author. Specializes in real-time image processing, edge AI across Jetson, Raspberry Pi, Hailo, Axelera, ARM. Full-stack CV/DL: model training → fine-tuning → deployment → API integration.\n\n**Metrics:** 21 publications (3 patents, 2 books, 6 journals, 11 conferences · h-index 7 · i10-index 5 · LinkedIn 55K+ · Facebook 15K+\n\n---\n\n# Publications\n\n## Patents\n\n### 1. Face Image Augmentation — WO 2021/060971 A1\n\nGenerates realistic face images from surveillance video using GANs. Captures faces from multiple angles, augments through data transformation..." },{ "id": "/notes/pubs/patents/vehicle-detection/", "title": "Moving Vehicle Detection WO2021107761A1", "url": "/notes/pubs/patents/vehicle-detection/", "tags": ["patent","vehicle-detection","image-processing","traffic-surveillance"], "hashtags": "#patent #vehicledetection #imageprocessing #trafficsurveillance", "category": "hub", "body": "> **Moving Vehicle Detection WO2021107761A1** — Image processing method for detecting moving vehicles using CLAHE, Sobel, and geometric filtering. — https://www.pirahansiah.com/notes/pubs/patents/vehicle-detection/\n\nspotify https://podcasters.spotify.com/pod/show/pirahansiah/episodes/A-METHOD-FOR-DETECTING-A-MOVING-VEHICLE-WO2021107761A1-e2pk5mj\n\n\n\n\n \n Your browser does not support the audio element.\n\n\nPDF Download https://patentimages.storage.googleapis.com/16/41/83/2576e20c4a0af5/WO2021107761A1.pdf \n\n\n{% if page.extname == \"vehicle-detection.md\" %}\n !A METHOD FOR DETECTING A MOVING VEHICLE WO2021107761A1 /notes/pubs/patents/vehicle-detection.png\n{% else %}\n \n{% endif %}\n\n\n\nPatent Summary:\n\n\nField of Invention:\n\nThis patent relates to a method for detecting moving vehicles, specifically in the field of traffic surveillance and law enforcement. It aims to improve vehicle detection accuracy under poor lighting conditions.\n\nBackground:\n\nExisting methods for detecting moving vehicles, such as edge processing and comparison between target and reference images, struggle under low light or bad weather conditions. This patent seeks to overcome these challenges.\n\nSummary of Invention:\n\nThe patented method enhances the detection of moving vehicles by processing video streams, focusing on improving the illumination and edge detection of the vehicle in the captured image. It involves:\n\n\t1.\tGrabbing an initial image from a video stream.\n\t2.\tEnhancing illumination of the image using Contrast Limited Adaptive Histogram Equalization (CLAHE if necessary.\n\t3.\tEnhancing edges using S..." },{ "id": "/notes/wiki/", "title": "All Pages — pirahansiah.com", "url": "/notes/wiki/", "tags": [], "hashtags": "", "category": "hub", "body": "\n\n\n\n All Pages\n From pirahansiah.com, the free knowledge base\n\n \n \n \n\n \n {{ site.pages size }} pages\n 3 patents\n 6 journals\n 11 papers\n 7 books\n 19 courses\n \n\n \n Contents\n \n Home\n Computer Vision\n AI & LLMs\n CUDA & GPU\n Programming\n Optimization\n Business & Career\n PKM & Knowledge\n Patents\n Journals\n Conference Papers\n Books\n Keynotes\n Courses\n Presentations\n \n \n\n \n Home\n \n Dr. Farshid Pirahansiah — Main landing page\n Computer Vision — CV overview and research\n \n \n\n \n Computer Vision\n \n 3D Vision & Multi-Camera Systems\n Optical Flow AI\n Multi-Camera Systems\n CV Coaching Roadmap\n Product — Embedded CV & Edge AI\n Computer Vision Overview\n \n \n\n \n AI & LLMs\n \n Advanced LLM Concepts New\n Orchestrating AI Agents\n Blog: AI & LLMs\n Avatar Generator\n \n \n\n \n CUDA & GPU\n \n Numba JIT Tutorial\n PyCUDA Kernels\n CUDA in VS Code\n MLX, CoreML & Metal\n \n \n\n \n Programming\n \n C++ Quick Reference\n Python Configuration\n Developer Tools & Setup\n Shell & Vim Reference\n \n \n\n \n Optimization\n \n CV, DL & ML Optimization\n Prompt Engineering\n \n \n\n \n Business & Career\n \n Startup Guide — Edge AI business in Germany\n SEO for LLMs\n Top LinkedIn Posts 2024\n \n \n\n \n PKM & Knowledge\n \n Content Hub\n Curated Links\n Use..." }]