Research map / 3D vision / 3D generation
SignSparK: Efficient Multilingual Sign Language Production via Sparse Keyframe Learning
Jian He Low, Alexandre Symeonidis-Herzig, Maksym Ivashechkin, Ozge Mercanoglu Sincan, 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 |
|---|---|---|---|
| remove 2x temporal downsampling (ablation) | FAST · MeineDGS segmentation | F1 / IoU: F1 +1%, IoU +1.8% | Improved |
| use dual-stream (one stream per hand) instead of single-stream (ablation) | FAST · MeineDGS segmentation | F1 / IoU | Improved |
| add temporal convolution layers (ablation) | FAST · MeineDGS segmentation | F1 / IoU | Improved |
| use unimodal WiLoR hand features instead of multimodal inputs (ablation) | FAST · MeineDGS segmentation | F1 | No clear effect |
| add reconstruction-based formulation to CFM (ablation) | SignSparK · Single-step generation | pose error: 3% gain in single-step generation | Improved |
| use segment-based keyframes instead of random sampling (ablation) | SignSparK · SLP keyframe selection | JPE | Improved |
| add random masking augmentation on segment keyframes (ablation) | SignSparK · SLP keyframe selection | JPE | Hurt |
| add reconstruction loss to CFM loss (ablation) | SignSparK · SLP loss configuration | JPE | Improved |
| add language identifier token (ablation) | SignSparK · Multilingual training with BOBSL | error rate | Improved |
| add BOBSL data without language identifier (ablation) | SignSparK · Multilingual training | error rate | Hurt |
| replace SLERP interpolation with learned SignSparK | KF2P · Phoenix-2014T KF2P | back-translation BLEU-4: 9.18 -> 11.70 | Improved |
| replace Sign Stitcher with SignSparK | Sign stitching · Phoenix-2014T | DTW-JPE: 57.20 -> 9.01 | Improved |
Closest papers by meaning
- Signing at Scale: Learning to Co-Articulate Signs for Large-Scale Photo-Realistic Sign Language Production (CVPR 2022 · similarity 0.91)
- SignPR: A Progressive Vector-Quantized Diffusion Framework for Sign Language Production (CVPR 2026 · similarity 0.90)
- SignAvatars: A Large-scale 3D Sign Language Holistic Motion Dataset and Benchmark (ECCV 2024 · similarity 0.90)
- Advanced Sign Language Video Generation with Compressed and Quantized Multi-Condition Tokenization (NeurIPS 2025 · similarity 0.88)
- Lost in Translation, Found in Embeddings: Sign Language Translation and Alignment (NeurIPS 2026 · similarity 0.88)
- Sign-IDD: Iconicity Disentangled Diffusion for Sign Language Production (AAAI 2025 · 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.
