Research map / Image generation / Autoregressive and latent generation
Focal-General Diffusion Model with Semantic Consistent Guidance for Sign Language Production
Yiheng Yu, Sheng Liu, Yuan Feng, Zhelun Jin, Yining Jiang, Min Xu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the Focal stage (ASGCN + TCN) to the General-only baseline (ablation) | FGDM · PHOENIX14T DEV | WER: 81.00 -> 76.57 | Improved |
| add Semantic Consistent Guidance (SCG) to the General-only baseline (ablation) | FGDM · PHOENIX14T DEV | WER: 81.00 -> 78.92 | Improved |
| combine Focal stage and SCG (ablation) | FGDM · PHOENIX14T TEST | WER: 79.76 -> 70.70 | Improved |
| add contextual correlation term A_a to ASGCN (ablation) | ASGCN · PHOENIX14T DEV | BLEU-1: 23.29 -> 28.09 | Improved |
| add skeletal topology term A_b to ASGCN (ablation) | ASGCN · PHOENIX14T DEV | BLEU-1: 26.44 -> 28.09 | Improved |
| add semantic mask M to ASGCN (ablation) | ASGCN · PHOENIX14T DEV | WER: 77.66 -> 76.57 | Improved |
| increase Focal layers from 3 to 4 (ablation) | FGDM · PHOENIX14T DEV | WER: 72.22 -> 72.35 | No clear effect |
| increase Focal layers from 1 to 3 (ablation) | FGDM · PHOENIX14T DEV | WER: 76.27 -> 72.22 | Improved |
| use 2 General layers instead of 1 or 3 (ablation) | FGDM · PHOENIX14T DEV | WER: 79.13 (L2=1), 73.79 (L2=3) -> 72.22 (L2=2) | Improved |
| FGDM versus strongest diffusion baseline Sign-IDD | FGDM · PHOENIX14T TEST | WER: 79.15 -> 70.70 | Improved |
| FGDM versus Sign-IDD on harder split | FGDM · USTC-CSL TEST (Split-II) | BLEU-3: 2.68 -> 15.71 | Improved |
Closest papers by meaning
- G2P-DDM: Generating Sign Pose Sequence from Gloss Sequence with Discrete Diffusion Model (AAAI 2024 · similarity 0.91)
- Discrete to Continuous: Generating Smooth Transition Poses from Sign Language Observations (CVPR 2025 · similarity 0.88)
- Sign-IDD: Iconicity Disentangled Diffusion for Sign Language Production (AAAI 2025 · similarity 0.88)
- Bridging Sign and Spoken Languages: Pseudo Gloss Generation for Sign Language Translation (NeurIPS 2025 · similarity 0.88)
- Sign2GPT: Leveraging Large Language Models for Gloss-Free Sign Language Translation (ICLR 2024 · similarity 0.88)
- Scaling Sign Language Translation (NeurIPS 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
