SLCC v1 metadata release

Resources: Demo · Paper [arXiv] · Code [GitHub] · Model weights [HuggingFace]

SLCC v1 contains the reviewed chessboard annotations used by the ChessQueries paper: metadata for 2,174 frames from 20 live broadcasts of the Grand Chess Tour 2026.

Try it online with the ChessQueries board recognition demo. The website uses a ChessQueries model further trained on proprietary data.

24 frames from the 20 source livestreams of the Saint Louis Chess Club.

The 152 games from which frames were labeled are assigned wholly to one split to prevent data leakage between train / validation / test.

Reconstruct the dataset

This repository contains metadata only: no videos, frames, crops, thumbnails, or learned descriptors. The GitHub code repository contains the pinned reconstruction and validation workflow. Users will need to obtain source videos directly from YouTube through the provided scripts.

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}
}

Rights and source availability

The project-authored annotations are licensed under CC BY-NC 4.0. The underlying broadcasts and decoded pixels remain third-party material; this release redistributes none of them and grants no rights over them. Users are responsible for complying with source-host terms and applicable law.

Contact

Questions and enquiries: contact "at" chessq.org