Computer Vision Playground

A curated collection of small, production-minded computer vision experiments.

Focus

• Detection, segmentation, pose estimation, tracking
• Real-time video pipelines
• GPU optimisation and benchmarking (ONNX / TensorRT where relevant)

Structure

• projects/ — focused experiments with their own README + runnable code
• tools/ — reusable benchmarking and helper scripts
• docs/assets/ — small curated demos for READMEs

How to run

Each project folder contains its own instructions.

Benchmarks

Projects include:

• FPS / latency notes
• CPU vs GPU comparisons when available

License

MIT

Projects

01 Video Pipelines

• Async Reader + FPS Baseline — sync vs async decode benchmark, structured JSON metrics

02 Object Detection

• YOLO26 Baseline — YOLO26n boat detection on video, COCO class filter, annotated video output

03 Pose Estimation

• Pose Baseline (PyTorch) — torchvision Keypoint RCNN, keypoint extraction + overlay

04 Tracking

• RAFT Boat Tracker — YOLO26n detections + RAFT optical flow propagation, IoU-based track matching, live preview window

More projects coming:

• Instance Segmentation
• Optimisation