This article shows how to recognize chess positions from images and run the entire pipeline directly in the browser.
The approach is to split the board into 64 squares, classify each square independently, and then reconstruct the position for analysis.
What’s covered:
- Creating a custom dataset of chess diagrams
- Learning representations with SimCLR and Autoencoder
- Training a CNN to classify chess pieces
Running preprocessing and inference client-side with Pyodide and ONNX Runtime Web
Top comments (0)