Research map / Image generation / Fast diffusion sampling
Semi-Implicit Denoising Diffusion Models (SIDDMs)
yanwu xu, Mingming Gong, Shaoan Xie, Wei Wei, Matthias Grundmann, kayhan Batmanghelich, Tingbo Hou
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 |
|---|---|---|---|
| Decompose denoising matching into adversarial marginal matching and explicit AFD conditional matching (ablation) | SIDDMs · CIFAR10 | FID: 77.15 -> 2.24 | Improved |
| Add adversarial marginal matching alongside the AFD term (ablation) | SIDDMs · CIFAR10 | FID: 41.27 -> 2.24 | Improved |
| Add the discriminator reconstruction regularizer (ablation) | SIDDMs · CIFAR10 | FID: 3.20 -> 2.24 | Improved |
| Use 2 diffusion steps instead of 1 (ablation) | SIDDMs · CIFAR10 | FID: 27.39 -> 3.17 | Improved |
| Use 4 diffusion steps instead of 2 (ablation) | SIDDMs · CIFAR10 | FID: 3.17 -> 2.24 | Improved |
| Use 16 diffusion steps instead of 4 (ablation) | SIDDMs · CIFAR10 | FID: 2.24 -> 2.15 | Improved |
| Increase diffusion steps from 16 to 32 (ablation) | SIDDMs · CIFAR10 | FID: 2.15 -> 2.46 | Hurt |
| Vary AFD weight while retaining both objective terms (ablation) | SIDDMs · CIFAR10 | FID: 2.24; 2.63; 2.55 | No clear effect |
| Use the semi-implicit objective instead of DDGAN training | SIDDMs and DDGANs · CIFAR10 | FID: 3.75 -> 2.24 | Improved |
| Use the semi-implicit objective instead of DDGAN training | SIDDMs and DDGANs · CelebA-HQ-256 | FID: 7.64 -> 7.37 | Improved |
| Use the semi-implicit objective instead of DDGAN training | SIDDMs and DDGANs · ImageNet1000-64 | FID: 20.63 -> 3.13 | Improved |
| Use SIDDMs rather than the best listed diffusion baseline | SIDDMs and ADM · ImageNet1000-64 | FID: 2.07 -> 3.13 | Hurt |
Closest papers by meaning
- Directly Denoising Diffusion Models (ICML 2024 · similarity 0.91)
- You Only Sample Once: Taming One-Step Text-to-Image Synthesis by Self-Cooperative Diffusion GANs (ICLR 2025 · similarity 0.90)
- Denoising Diffusion Bridge Models (ICLR 2024 · similarity 0.90)
- UDPM: Upsampling Diffusion Probabilistic Models (NeurIPS 2024 · similarity 0.90)
- Revisiting Diffusion Models: From Generative Pre-training to One-Step Generation (ICML 2025 · similarity 0.90)
- Interpreting and Improving Diffusion Models from an Optimization Perspective (ICML 2024 · similarity 0.90)
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.
