ChessQueries model weights

Resources: Paper [arXiv] · Code [GitHub] · SLCC dataset annotations [HuggingFace]

ChessQueries can handle four datasets with various challenges.

Release weights for ChessQueries, a real-life chessboard recognizer using a DINOv2 ViT-L/14 encoder and a square-query decoder. The model takes 644 × 644 inputs and was jointly trained on ChessReD, ChessCog, and SLCC.

Files

• chessqueries-vitL14-644-joint.safetensors — recommended, tensor-only inference weights; SHA-256 6151bdd98fbe25f32080c097eba7ae75808b5615160e4867240931085fc762e5.
• chessqueries-vitL14-644-joint.ckpt — original full PyTorch Lightning checkpoint; SHA-256 958e30d0a982873cfaef7fd3267489c8e554a8b2e75a9fb8441a511cd5a0ff18.

Setup, inference, evaluation, and reproduction instructions are in the code repository.

Citation

@misc{seytre2026chessqueries,
  title         = {ChessQueries: Toward Better Chess Board Recognition},
  author        = {Seytre, Jo{\"e}l},
  year          = {2026},
  eprint        = {2608.30762},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  doi           = {10.48550/arXiv.2608.30762},
  url           = {https://arxiv.org/abs/2608.30762}
}

License

The ChessQueries model weights are licensed under the PolyForm Noncommercial License 1.0.0. They may be used, modified, and redistributed for permitted noncommercial purposes; commercial use is not licensed. Third-party components, including the DINOv2 encoder, remain subject to their respective upstream terms.