Research map / Image generation / Fast diffusion sampling
One-step Optimal Transport via Regularized Distribution Matching Distillation
Denis Rakitin, Ivan Shchekotov, Viacheslav Meshchaninov, Dmitry Vetrov
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 |
|---|---|---|---|
| use LPIPS instead of L2 as transport cost (ablation) | RDMD · AFHQv2 64x64 Wild->Cat | FID: 17.31 -> 17.17 | No clear effect |
| use LPIPS instead of L2 as transport cost (ablation) | RDMD · AFHQv2 64x64 Wild->Cat | L2: 14.71 -> 19.49 | Hurt |
| use LPIPS instead of L2 cost for color inversion task (ablation) | RDMD · Inverted-MNIST to MNIST | semantic correctness | Improved |
| replace adversarial OT training with diffusion distribution matching (RDMD vs DIOTM) | OT translation · CelebA 64x64 M2F (5k samples) | FID: 31.34 -> 20.99 | Improved |
| replace adversarial OT training with RDMD (vs ASBM) | OT translation · CelebA 64x64 M2F (5k samples) | FID: 43.97 -> 20.99 | Improved |
| reduce training data to 5k samples (RDMD) | RDMD · CelebA 64x64 M2F | FID: 10.36 -> 20.99 | Hurt |
| reduce training data to 5k samples (ASBM) | ASBM · CelebA 64x64 M2F | FID: 23.06 -> 43.97 | Hurt |
| reduce training data to 5k samples (DIOTM) | DIOTM · CelebA 64x64 M2F | FID: 15.81 -> 31.34 | Hurt |
| RDMD vs strongest OT baseline DIOTM | one-step translation · AFHQv2 64x64 Wild->Cat | LPIPS: 0.417 -> 0.352 | Improved |
| RDMD vs DDIB on realism | translation · CelebA 128x128 Male->Female | FID: 5.39 -> 9.30 | Hurt |
| RDMD vs EGSDE† at 256x256 | translation · AFHQv2 256x256 Cat->Dog | FID: 51.04 -> 44.32 | Improved |
| RDMD vs EGSDE† at 256x256 | translation · AFHQv2 256x256 Cat->Dog | L2: 62.06 -> 64.44 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- EGSDE† — EGSDE: Unpaired Image-to-Image Translation via Energy-Guided Stochastic Differential Equations (NeurIPS 2022) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Efficient One-to-many Domain Translation via Diffusive Entropic Optimal Transport (NeurIPS 2026 · similarity 0.89)
- Adversarial Distribution Matching for Diffusion Distillation Towards Efficient Image and Video Synthesis (ICCV 2025 · similarity 0.89)
- OT-ALD: Aligning Latent Distributions with Optimal Transport for Accelerated Image-to-Image Translation (AAAI 2026 · similarity 0.88)
- One-step Diffusion with Distribution Matching Distillation (CVPR 2024 · similarity 0.87)
- Improved Distribution Matching Distillation for Fast Image Synthesis (NeurIPS 2024 · similarity 0.87)
- Continuous-Time Distribution Matching for Few-Step Diffusion Distillation (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.
