Research map / Video / Video generation
Content-Aware Dynamic Patchification for Efficient Video Diffusion
Sheng Li, Connelly Barnes, Mamshad Nayeem Rizve, Hongwu Peng, Zhengang Li, Ohi Dibua, Alireza Ganjdanesh, Xulong Tang, Yan Kang, Yifan Gong
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 attention map guidance loss to router training (20% reduction) (ablation) | DynaPatch · VBench, 20% token reduction | Total Score: 82.28 -> 83.56 | Improved |
| add attention map guidance loss to router training (30% reduction) (ablation) | DynaPatch · VBench, 30% token reduction | Total Score: 82.05 -> 83.42 | Improved |
| add attention map guidance loss to router training (40% reduction) (ablation) | DynaPatch · VBench, 40% token reduction | Total Score: 80.54 -> 82.19 | Improved |
| learned jointly-trained content-aware router vs entropy-based router (D2iT) at 30% reduction | DynaPatch · VBench, 30% token reduction | Total Score: 81.08 -> 83.42 | Improved |
| region-wise spatiotemporal routing vs timestep-level routing (FlexiDiT) at 40% reduction | DynaPatch · VBench, 40% token reduction | Total Score: 79.67 -> 82.19 | Improved |
| patchification at token creation vs token pruning (SPViT) at 30% reduction | DynaPatch · VBench, 30% token reduction | Total Score: 79.20 -> 83.42 | Improved |
| add larger patch size (2,8,8) to candidate set (ablation) | DynaPatch · patch size candidates | temporal consistency | Hurt |
| reduce tokens 30% with dynamic patchification vs uniform finest patch | DynaPatch · VBench | Total Score: 83.61 -> 83.42 | No clear effect |
Closest papers by meaning
- Hierarchical Patch Diffusion Models for High-Resolution Video Generation (CVPR 2024 · similarity 0.90)
- PatchVSR: Breaking Video Diffusion Resolution Limits with Patch-wise Video Super-Resolution (CVPR 2025 · similarity 0.90)
- Training-Free Efficient Video Generation via Dynamic Token Carving (NeurIPS 2025 · similarity 0.89)
- SparseDiT: Token Sparsification for Efficient Diffusion Transformer (NeurIPS 2025 · similarity 0.88)
- From Slow Bidirectional to Fast Autoregressive Video Diffusion Models (CVPR 2025 · similarity 0.88)
- MotionAura: Generating High-Quality and Motion Consistent Videos using Discrete Diffusion (ICLR 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.
