Research map / Image generation / Fast diffusion sampling
Distilling Diffusion Models into Conditional GANs
MinGuk Kang, Richard Zhang, Connelly Barnes, Sylvain Paris, Suha Kwak, Jaesik Park, Eli Shechtman, Jun-Yan Zhu, Taesung Park
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 MSE with E-LatentLPIPS as ODE distillation loss (ablation) | ODE distillation · SD-CFG-3, COCO2014 | FID: 110.55 -> 22.95 | Improved |
| add ensembled augmentations to LatentLPIPS (ablation) | ODE distillation · SD-CFG-3, COCO2014 | FID: 67.17 -> 22.95 | Improved |
| use E-LatentLPIPS in latent space instead of pixel LPIPS (ablation) | ODE distillation · SD-CFG-3, COCO2014 | FID: 25.94 -> 22.95 | Improved |
| increase batch size from 256 to 2048 (ablation) | ODE distillation with E-LatentLPIPS · SD-CFG-3, COCO2014 | FID: 22.95 -> 14.72 | Improved |
| add pre-trained diffusion discriminator instead of GigaGAN discriminator (ablation) | Diffusion2GAN · SD-CFG-3, COCO2014 | FID-30k: 13.97 -> 12.04 | Improved |
| add single-sample R1 regularization (ablation) | Diffusion2GAN discriminator · SD-CFG-3, COCO2014 | FID-30k: 11.97 -> 10.60 | Improved |
| add multi-scale in-and-out discriminator (ablation) | Diffusion2GAN · SD-CFG-3, COCO2014 | FID-30k: 10.60 -> 9.58 | Improved |
| add mix-and-match augmentation (ablation) | Diffusion2GAN · SD-CFG-3, COCO2014 | FID-30k: 9.58 -> 9.45 | Improved |
| add z-conditioning to the discriminator (ablation) | Diffusion2GAN discriminator · SD-CFG-3, COCO2014 | FID-30k: 12.04 -> 11.97 | No clear effect |
| add conditional GAN training to ODE distillation (ablation) | ODE distillation · COCO2014 | FID-30k: 15.94 -> 9.29 | Improved |
| use deterministic ODE pairs vs stochastic or re-wired pairs (ablation) | ODE distillation · EDM CIFAR10, 50k pairs | FID: FID over 200 -> 8.51 | Improved |
| scale up ODE pair dataset (ablation) | ODE distillation (LPIPS) · CIFAR10 | FID-50k: 8.51 -> 3.16 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- UFOGen — UFOGen: You Forward Once Large Scale Text-to-Image Generation via Diffusion GANs (CVPR 2024) · better on 2 of 2 shared results
- InstaFlow-0.9B — InstaFlow: One Step is Enough for High-Quality Diffusion-Based Text-to-Image Generation (ICLR 2024) · 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
- Revisiting Diffusion Models: From Generative Pre-training to One-Step Generation (ICML 2025 · similarity 0.92)
- Unlocking Dataset Distillation with Diffusion Models (NeurIPS 2025 · similarity 0.91)
- E$^2$GAN: Efficient Training of Efficient GANs for Image-to-Image Translation (ICML 2024 · similarity 0.89)
- Neural Diffusion Models (ICML 2024 · similarity 0.88)
- Structure-Guided Adversarial Training of Diffusion Models (CVPR 2024 · similarity 0.88)
- Unified Generation, Reconstruction, and Representation: Generalized Diffusion with Adaptive Latent Encoding-Decoding (ICML 2024 · 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.
