Research map / Video / Video generation
SignPR: A Progressive Vector-Quantized Diffusion Framework for Sign Language Production
Xiao Liu, Shiwei Gan, Yafeng Yin, Bowen Guo, Zhiwei Jiang, Shunmei Meng, Lei Xie, Sanglu Lu
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 |
|---|---|---|---|
| use SignPR (no glosses) instead of gloss-based Sign-IDD | sign pose generation · PHOENIX-14T test | MPJPE: 47.19 -> 23.04 | Improved |
| add regional codebooks to the semantic codebook in S-VQVAE (ablation) | S-VQVAE · PHOENIX-14T reconstruction | rROUGE: 36.69 -> 39.51 | Improved |
| share one regional codebook across regions instead of separate codebooks (ablation) | S-VQVAE · PHOENIX-14T reconstruction | rROUGE: 39.51 -> 37.54 | Hurt |
| add structural consistency loss Lcons (ablation) | SignPR · PHOENIX-14T test | ROUGE: 24.35 -> 32.86 | Improved |
| use GCN encoder instead of Transformer encoder (ablation) | S-VQVAE · PHOENIX-14T test | MPJPE: 28.07 -> 23.04 | Improved |
| add regional diffusion on top of semantic diffusion (ablation) | S-Diffusion · PHOENIX-14T test | ROUGE: 22.53 -> 32.86 | Improved |
| add semantic diffusion on top of regional diffusion (ablation) | S-Diffusion · PHOENIX-14T test | ROUGE: 23.90 -> 32.86 | Improved |
| use separate regional codebooks instead of a shared regional codebook in diffusion (ablation) | S-Diffusion · PHOENIX-14T test | ROUGE: 27.49 -> 32.86 | Improved |
| condition regional diffusion on semantic pose in addition to text, using cross-attention (ablation) | regional diffusion · PHOENIX-14T test | ROUGE: 23.90 (text only) -> 32.86 | Improved |
| fuse text and semantic pose by cross-attention instead of concatenation (ablation) | regional diffusion · PHOENIX-14T test | ROUGE: 26.42 -> 32.86 | Improved |
| add block-wise causal inference (InferRef, K=8) at inference time (ablation) | SignPR · PHOENIX-14T test | ROUGE: 31.50 -> 32.86 | Improved |
| use smaller blocks (K=8 instead of unblocked) in InferRef (ablation) | SignPR · PHOENIX-14T test | seconds per video: 1.47 -> 2.45 | Hurt |
Closest papers by meaning
- Focal-General Diffusion Model with Semantic Consistent Guidance for Sign Language Production (CVPR 2026 · similarity 0.92)
- G2P-DDM: Generating Sign Pose Sequence from Gloss Sequence with Discrete Diffusion Model (AAAI 2024 · similarity 0.89)
- Discrete to Continuous: Generating Smooth Transition Poses from Sign Language Observations (CVPR 2025 · similarity 0.89)
- Neural Sign Actors: A Diffusion Model for 3D Sign Language Production from Text (CVPR 2024 · similarity 0.89)
- Advanced Sign Language Video Generation with Compressed and Quantized Multi-Condition Tokenization (NeurIPS 2025 · similarity 0.89)
- Sign-IDD: Iconicity Disentangled Diffusion for Sign Language Production (AAAI 2025 · similarity 0.89)
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.
