SignSparK Data

Prebuilt LMDB datasets for SignSparK: Efficient Multilingual Sign Language Production via Sparse Keyframe Learning (ECCV 2026). Contains SMPL-X 6D pose features, spoken-language translations, glosses, and per-frame segment annotations for CSL-Daily, How2Sign and PHOENIX-2014T, ready to train/sample with the SignSparK code.

• 📄 Paper: https://arxiv.org/abs/2603.10446
• 🌐 Project page: https://cogvis-cvssp.github.io/papers/signspark/

Contents

train/  CSL-Daily_reopt_train.lmdb   How2Sign_reopt_train.lmdb   PHOENIX-2014T_reopt_train.lmdb
dev/    CSL-Daily_reopt_dev.lmdb     How2Sign_reopt_dev.lmdb     PHOENIX-2014T_reopt_dev.lmdb
test/   CSL-Daily_reopt_test.lmdb    How2Sign_reopt_test.lmdb    PHOENIX-2014T_reopt_test.lmdb

Each .lmdb holds one __meta__ key (clip index) plus one in-memory .npz blob per clip with: language, translation, gloss, segment (per-frame {0,1,2}), and left/right/body/face_features (6D rotations). Full schema in the repo's DATA.md.

Usage

python tools/download_data.py --datasets CSL-Daily How2Sign PHOENIX-2014T --dest ./data
export DATA_ROOT=$(pwd)/data
python tools/inspect_lmdb.py ${DATA_ROOT}/lmdb/train/CSL-Daily_reopt_train.lmdb

License & terms

These are SMPL-X pose features derived from the CSL-Daily, How2Sign and PHOENIX-2014T corpora and are released for non-commercial research use under the terms of the original datasets. Obtain and comply with each source dataset's license:

• CSL-Daily: https://ustc-slr.github.io/datasets/2021_csl_daily/
• How2Sign: https://how2sign.github.io/
• PHOENIX-2014T: https://www-i6.informatik.rwth-aachen.de/~koller/RWTH-PHOENIX-2014-T/

Citation

@inproceedings{low2026signspark,
  title={SignSparK: Efficient Multilingual Sign Language Production via Sparse Keyframe Learning},
  author={Low, Jianhe and Symeonidis-Herzig, Alexandre and Ivashechkin, Maksym and Sincan, Ozge Mercanoglu and Bowden, Richard},
  booktitle={European Conference on Computer Vision},
  pages={648--670},
  year={2026},
  organization={Springer}
}