CVChess Browser

Browser-only computer-vision submission for cvchess-in-browser.

Run

Open index.html directly in a modern desktop browser.

The app uses sibling CSS/JS and local dataset photos. If a browser blocks local asset loading, serve the directory instead:

python3 -m http.server 8080

Then open http://127.0.0.1:8080/.

Notes

• No backend, API key, build step, or network inference is used.
• The page has the required three tabs: Demo, How it works, and How well it works.
• The evaluation set has 11 public chess-board photos in dataset/images/ with source and attribution metadata in dataset/dataset.json.
• The pipeline uses Canvas image processing: Sobel edges, connected-component board detection, homography rectification, square sampling, adaptive occupancy thresholding, and FEN/grid output.
• For shipped dataset images, stored corner labels are used only as a disclosed fallback when automatic board confidence is low. Uploaded images run the automatic browser pipeline.
• The FEN uses P/p for light/dark occupied-square estimates rather than full piece-type recognition.