Research map / Image generation / Fast diffusion sampling
Optimizing Few-Step Generation with Adaptive Matching Distillation
Lichen Bai, zikai zhou, Shitong Shao, Wenliang Zhong, Shuo Yang, Shuo Chen, Bojun Cheng, Zeke Xie
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 DMD2 with AMD (reward-aware adaptive distillation) | SDXL 4-step student · COCO2014-val 10k | HPSv2: 30.64 -> 31.25 | Improved |
| replace DMD2 with AMD | SDXL 4-step student · COCO2014-val 10k | ImageReward: 71.01 -> 88.37 | Improved |
| replace DMD2 with AMD | SDXL 4-step student · GenEval | overall score: 0.51 -> 0.57 | Improved |
| replace standard DMD with AMD | SiT-XL/2 · ImageNet 256x256 | FID: 3.5573 -> 3.4690 | Improved |
| replace DMDR (reward RL steering) with AMD | SiT-XL/2 · ImageNet 256x256 | FID: 9.6341 -> 3.4690 | Improved |
| replace standard DMD with AMD | SiT-XL/2 · ImageNet 256x256 | IS: 314.42 -> 316.02 | No clear effect |
| replace LongLive baseline with AMD | Wan2.1-1.3B streaming · VBench | Motion Quality: 35.51 -> 59.26 | Improved |
| replace LongLive baseline with AMD | Wan2.1-1.3B streaming · VBench total score | Total: 81.42 -> 82.21 | Improved |
| replace LongLive baseline with AMD | Wan2.1-1.3B streaming · VBench preference metrics | Text Alignment: 107.98 -> 100.79 | Hurt |
| replace LongLive baseline with AMD | Wan2.1-1.3B streaming · VBench semantic score | Semantic: 70.38 -> 69.91 | Hurt |
| add Dynamic Score Adaptation to baseline DMD (ablation) | SiT-XL/2 DMD baseline · ImageNet 256x256 | FID/IS | Improved |
| add Repulsive Landscape Sharpening (ablation) | SiT-XL/2 DMD with Dynamic Adapt · ImageNet 256x256 | FID/IS | Improved |
Closest papers by meaning
- Continuous-Time Distribution Matching for Few-Step Diffusion Distillation (NeurIPS 2026 · similarity 0.89)
- Diff-Instruct with Diffused Reward: Towards Principled One-step Generator RL (NeurIPS 2026 · similarity 0.89)
- Improved Distribution Matching Distillation for Fast Image Synthesis (NeurIPS 2024 · similarity 0.89)
- Distribution Matching Distillation without Fake Score Network (NeurIPS 2026 · similarity 0.88)
- CoDMD: Copula-aware Distribution Matching Distillation for Fast Video Generation (NeurIPS 2026 · similarity 0.88)
- GDMD: Guiding Distribution Matching Distillation with Gradient-Based Reinforcement Learning (NeurIPS 2026 · 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.
