Research map / Learning theory and algorithms / Optimal transport
OTCache: Optimal Transport for Geometry-Aware Caching in Diffusion Models
Huanlin Gao, Fang Zhao, Qiang Hui, Fuyuan Shi, Shaoan Zhao, Yantao Li, Chao Tan, Ting Lu, Yuren You, Kai Wang, Shiguo Lian
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 |
|---|---|---|---|
| replace MeanCache schedule with OTCache schedule (B=15) | MeanCache · FLUX.1 [dev] | LPIPS: 0.142 -> 0.126 | Improved |
| replace MeanCache schedule with OTCache schedule (B=10) | MeanCache · FLUX.1 [dev] | speedup: 4.12x -> 4.50x | Improved |
| replace MeanCache with OTCache (B=15) | MeanCache · Qwen-Image | LPIPS: 0.075 -> 0.069 | Improved |
| replace MeanCache with OTCache (B=10) | MeanCache · Qwen-Image | LPIPS: 0.236 -> 0.171 | Improved |
| replace MeanCache with OTCache (B=12) | MeanCache · HunyuanVideo | LPIPS: 0.176 -> 0.162 | Improved |
| replace MeanCache with OTCache (B=10) | MeanCache · HunyuanVideo | VBench: 80.08% -> 80.37% | Improved |
| replace MeanCache with OTCache (B=15) | MeanCache · FLUX.1 [dev] | CLIP score: 31.244 -> 31.167 | No clear effect |
| use power-law warping rho=1.3 instead of uniform rho=1.0 (ablation) | OTCache · FLUX.1 [dev] (B=15) | LPIPS: 0.139 -> 0.126 | Improved |
| use power-law warping rho=1.3 instead of uniform rho=1.0 (ablation) | OTCache · FLUX.1 [dev] (B=15) | ImageReward: 1.022 -> 1.011 | Hurt |
| raise rho from 1.3 to 1.45 (ablation) | OTCache · FLUX.1 [dev] (B=15) | PSNR: 26.034 -> 25.798 | Hurt |
| use anchor budget 8 instead of 6 (ablation) | OTCache · FLUX.1 [dev] (B=10) | LPIPS: 0.301 -> 0.254 | Improved |
| use anchor budget 8 instead of 4 (ablation) | OTCache · FLUX.1 [dev] (B=10) | PSNR: 18.29 -> 20.22 | Improved |
Closest papers by meaning
- Cache Me if You Can: Accelerating Diffusion Models through Block Caching (CVPR 2024 · similarity 0.90)
- CacheQuant: Comprehensively Accelerated Diffusion Models (CVPR 2025 · similarity 0.89)
- Learning-to-Cache: Accelerating Diffusion Transformer via Layer Caching (NeurIPS 2024 · similarity 0.89)
- Masked Diffusion Models as Energy Minimization (NeurIPS 2025 · similarity 0.88)
- Accelerating Rectified Flow Models via Trajectory-Aware Caching (NeurIPS 2026 · similarity 0.88)
- DeepCache: Accelerating Diffusion Models for Free (CVPR 2024 · similarity 0.87)
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.
