DARSLP

Stage 2 DARSLPGenerator checkpoints for Disentangle and Regularize: Sign Language Production with Articulator-Based Disentanglement and Channel-Aware Regularization (Taşyürek et al., WACV 2026).

Code, Stage 1 DisentangledAE checkpoints, and full training/inference instructions: github.com/sumeyyemeryem/DARSLP

Files

File	Dataset
darslp_generator_phoenix.ckpt	PHOENIX-2014T
darslp_generator_csl.ckpt	CSL-Daily

Usage

from src.models.darslp_generator import DARSLPGenerator

model = DARSLPGenerator.load_from_checkpoint(
    "darslp_generator_phoenix.ckpt",
    ae_model=ae_model,       # frozen Stage 1 DisentangledAE, see GitHub repo
    text_vocab=vocab,
    cfg=cfg, args=args,
    num_joints=176, num_feats=3, pose_dim=80,
)

See predict_phoenix.py / predict_phoenix.py in the GitHub repo for the full working example.

Citation

@INPROCEEDINGS{11492489,
  author={Taşyürek, Sümeyye Meryem and Kızıltepe, Tuğçe and Keles, Hacer Yalim},
  booktitle={2026 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  title={Disentangle and Regularize: Sign Language Production with Articulator-Based Disentanglement and Channel-Aware Regularization},
  year={2026},
  pages={8458-8467},
  doi={10.1109/WACV61042.2026.00816}}