Research map / Architectures and efficiency / Attention and transformers
Continuous 3D Multi-Channel Sign Language Production via Progressive Transformers and Mixture Density Networks
Ben Saunders, Necati Cihan Camgöz, Richard Bowden
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 future prediction of 5 frames (ablation) | progressive transformer · PHOENIX14T Gloss to Pose (dev) | BLEU-4: 7.38 -> 11.30 | Improved |
| use Just Counter input instead of full skeleton pose (ablation) | progressive transformer · PHOENIX14T Gloss to Pose (dev) | BLEU-4: 7.38 -> 12.34 | Improved |
| add Gaussian noise augmentation r_n=5 (ablation) | progressive transformer · PHOENIX14T Gloss to Pose (dev) | BLEU-4: 7.38 -> 12.80 | Improved |
| raise noise rate r_n from 5 to 10 (ablation) | Gaussian noise augmentation · PHOENIX14T Gloss to Pose (dev) | BLEU-4: 12.80 -> 12.14 | Hurt |
| condition discriminator on source input (ablation) | adversarial training · PHOENIX14T Gloss to Pose (dev) | BLEU-4: 12.65 -> 13.13 | Improved |
| increase discriminator layers N from 3 to 6 (ablation) | conditional adversarial · PHOENIX14T Gloss to Pose (dev) | BLEU-4: 12.76 -> 13.13 | Improved |
| increase mixture components M from 1 to 4 (ablation) | MDN · PHOENIX14T Gloss to Pose (dev) | BLEU-4: 12.22 -> 13.14 | Improved |
| increase mixture components M from 4 to 50 (ablation) | MDN · PHOENIX14T Gloss to Pose (dev) | BLEU-4: 13.14 -> 11.15 | Hurt |
| combine MDN with adversarial loss (ablation) | MDN · PHOENIX14T Gloss to Pose (dev) | BLEU-4: 13.14 -> 12.88 | Hurt |
| combine MDN with adversarial loss (ablation) | MDN · PHOENIX14T Gloss to Pose (test) | BLEU-4: 11.94 -> 12.32 | Improved |
| add Gaussian noise augmentation (ablation) | progressive transformer · PHOENIX14T Text to Pose (dev) | BLEU-4: 7.30 -> 10.75 | Improved |
| use MDN instead of Gaussian noise baseline (ablation) | progressive transformer · PHOENIX14T Text to Pose (dev) | BLEU-4: 10.75 -> 11.54 | Improved |
Closest papers by meaning
- Progressive Transformers for End-to-End Sign Language Production (ECCV 2020 · similarity 0.96)
- Mixed SIGNals: Sign Language Production via a Mixture of Motion Primitives (ICCV 2021 · similarity 0.92)
- Neural Sign Actors: A Diffusion Model for 3D Sign Language Production from Text (CVPR 2024 · similarity 0.91)
- Scaling Sign Language Translation (NeurIPS 2024 · similarity 0.89)
- Lost in Translation, Found in Embeddings: Sign Language Translation and Alignment (NeurIPS 2026 · similarity 0.89)
- Conditional Variational Autoencoder for Sign Language Translation with Cross-Modal Alignment (AAAI 2024 · 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.
