Research map / Video / Video generation
SVG-EAR: Parameter-Free Linear Compensation for Sparse Video Generation via Error-aware Routing
Xuanyi Zhou, Qiuyang Mang, Shuo Yang, Haocheng Xi, Jintao Zhang, Huanzhi Mao, Joseph E Gonzalez, Kurt Keutzer, Ion Stoica, Alvin Cheung
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 error-aware routing on top of linear compensation (ablation) | SVG2 + compensation · Wan2.2 T2V | PSNR: 24.157 -> 24.995 | Improved |
| add linear compensation to top-p selection (ablation) | SVG2 · Wan2.2 T2V | PSNR: 23.556 -> 24.157 | Improved |
| use full SVG-EAR instead of SVG2 | SVG2 · Wan2.2 T2V | PSNR: 23.556 -> 24.995 | Improved |
| use SVG-EAR instead of SVG2 | SVG2 · HunyuanVideo T2V | PSNR: 29.445 -> 31.043 | Improved |
| use SVG-EAR instead of SVG2 | SVG2 · Wan2.2 I2V | PSNR: 27.668 -> 29.759 | Improved |
| use SVG-EAR instead of SVG2 | SVG2 · Wan2.2 T2V | density: 32.30% -> 25.95% | Improved |
| reduce query clusters from 300 to 100 (ablation) | SVG-EAR · Wan2.2 T2V | PSNR: 24.995 -> 23.862 | Hurt |
| reduce key clusters from 1000 to 500 (ablation) | SVG-EAR · Wan2.2 T2V | PSNR: 24.995 -> 24.451 | Hurt |
| reduce query clusters to 50 with 500 key clusters (ablation) | SVG-EAR · Wan2.2 T2V | PSNR: 24.995 -> 22.501 | Hurt |
| increase number of query centroids (ablation) | SVG-EAR attention · Wan2.2 T2V | attention MSE | Improved |
| add linear compensation to top-p selection without error-aware routing (ablation) | SVG2 · Wan2.2 T2V | attention MSE | No clear effect |
| use Turbo configuration (fewer centroids, lower budget) (ablation) | SVG-EAR · Wan2.2 T2V | PSNR: 24.995 -> 23.940 | Mixed |
Closest papers by meaning
- Sparse VideoGen2: Accelerate Video Generation with Sparse Attention via Semantic-Aware Permutation (NeurIPS 2025 · similarity 0.93)
- Sparse Video-Gen: Accelerating Video Diffusion Transformers with Spatial-Temporal Sparsity (ICML 2025 · similarity 0.93)
- iCATS: Fast Video Generation via Interaction-Aware Sparse Attention and Timestep-Adaptive Sparsity (NeurIPS 2026 · similarity 0.89)
- PISA: Piecewise Sparse Attention Is Wiser for Efficient Diffusion Transformers (NeurIPS 2026 · similarity 0.89)
- Faster Video Diffusion with Trainable Sparse Attention (NeurIPS 2025 · similarity 0.89)
- MonarchRT: Efficient Attention for Real-Time Video Generation (NeurIPS 2026 · 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.
