CV Brief · Sunday, 25 October 2026
CV Brief
Tutorials & Guides
LiDAR obstacle detection: 175k points to robust rail geometry
Metro hackathon solution demonstrates 3D LiDAR point cloud processing for track inspection, converting raw 175k point clouds into confirmed obstacles via robust geometric fitting. Practical walkthrough of production-grade LiDAR pipeline for infrastructure monitoring.
Read more →Remove road shadows without training: CPU-first filtering approach
Production pipeline for shadow removal using local statistics, filters, and constraints instead of neural networks. Single real road image demonstrates how classical CV techniques handle real-world lighting artifacts efficiently on CPU.
Read more →Industry & Deployments
CNN inductive bias and local connectivity: linear algebra foundations
Series deep-dive into mathematical intuition behind convolutional layers, their architectural constraints, and why they work for vision tasks. Essential theory for practitioners optimizing model design and understanding architecture trade-offs.
Read more →Reasoning models course: 11 hours PyTorch, inference scaling, distillation
Comprehensive video course covering model evaluation, inference-time compute scaling, reinforcement learning, and knowledge distillation with PyTorch implementations. Bridges theory and production-ready techniques for modern vision systems.
Read more →When setting up train/val/test splits: split by scene or location, not just randomly by image. Random splits from the same video = data leakage and falsely high validation accuracy.