Research map / Image generation / Fast diffusion sampling
Focusing on What Matters: Saliency-Harnessing Accurate Routing for Diffusion MoE
Haoyou Deng, Keyu Yan, Chaojie Mao, Xiang Wang, Yu Liu, Changxin Gao, Nong Sang
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 |
|---|---|---|---|
| route with predicted clean latent x0 from previous step (saliency-harnessing router) (ablation) | DiffMoE-B · ImageNet 256x256 (cfg=1.5) | FID50K: 8.03 -> 6.95 | Improved |
| add trajectory routing loss (ablation) | DiffMoE-B + saliency router · ImageNet 256x256 (cfg=1.5) | FID50K: 6.95 -> 6.66 | Improved |
| post-train with SharpMoE | DiffMoE-L · ImageNet 256x256 (cfg=1.5) | FID50K: 3.86 -> 3.10 | Improved |
| post-train with SharpMoE | EC-DiT-L · ImageNet 256x256 (cfg=1.5) | FID50K: 4.09 -> 3.27 | Improved |
| post-train with SharpMoE (no routing loss, uniform allocation) | TC-DiT-L · ImageNet 256x256 (cfg=1.5) | FID50K: 5.07 -> 3.72 | Improved |
| post-train with SharpMoE | DiffMoE-S · ImageNet 256x256 (cfg=1.0) | FID50K: 46.55 -> 45.52 | Improved |
| vary rollout steps T from 5 to 20 in full-trajectory training (ablation) | SharpMoE · ImageNet 256x256 | FID50K | No clear effect |
Closest papers by meaning
- ALTER: All-in-One Layer Pruning and Temporal Expert Routing for Efficient Diffusion Generation (NeurIPS 2025 · similarity 0.89)
- Expert Race: A Flexible Routing Strategy for Scaling Diffusion Transformer with Mixture of Experts (ICML 2025 · similarity 0.89)
- Diff-MoE: Diffusion Transformer with Time-Aware and Space-Adaptive Experts (ICML 2025 · similarity 0.88)
- Breaking the Uniformity Trap: Scaling Video Diffusion Model via SplitMoE (NeurIPS 2026 · similarity 0.88)
- Rethinking Cross-Layer Information Routing in Diffusion Transformer (NeurIPS 2026 · similarity 0.87)
- EC-DIT: Scaling Diffusion Transformers with Adaptive Expert-Choice Routing (ICLR 2025 · 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.
