Research map / Image generation / Fast diffusion sampling
Diffusion Models Beat GANs on Image Synthesis
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| add classifier guidance (scale 1.0) to conditional diffusion model (ablation) | ADM · ImageNet 256x256 | FID: 10.94 -> 4.59 | Helped |
| add classifier guidance (scale 10.0) to unconditional model (ablation) | ADM unconditional · ImageNet 256x256 | FID: 26.21 -> 12.00 | Helped |
| classifier guidance scale 1.0 on unconditional model (ablation) | ADM unconditional · ImageNet 256x256 | FID: 26.21 -> 33.03 | Hurt |
| classifier guidance (ablation) | ADM · ImageNet 256x256 | Recall: 0.63 -> 0.52 | Hurt |
| increase guidance scale from 1.0 to 10.0 on conditional model (ablation) | ADM-G · ImageNet 256x256 | FID: 4.59 -> 9.11 | Hurt |
| use AdaGN instead of addition + GroupNorm (ablation) | ADM · ImageNet 128x128 | FID: 15.08 -> 13.06 | Helped |
| use BigGAN residual blocks for up/downsampling (ablation) | UNet diffusion · ImageNet 128x128 | FID: -1.20 | Helped |
| multi-resolution attention (32,16,8) (ablation) | UNet diffusion · ImageNet 128x128 | FID: -0.72 | Helped |
| increase attention heads to 4 (ablation) | UNet diffusion · ImageNet 128x128 | FID: -0.54 | Helped |
| increase depth to 4 with 128 channels (ablation) | UNet diffusion · ImageNet 128x128 | FID: -0.21 | Helped |
| rescale residual connections by 1/sqrt(2) (ablation) | UNet diffusion · ImageNet 128x128 | FID: 0.16 | Hurt |
| combine all architecture changes (ablation) | UNet diffusion · ImageNet 128x128 | FID: -3.14 | Helped |
Closest papers by meaning
- Revisiting Diffusion Models: From Generative Pre-training to One-Step Generation (ICML 2025 · similarity 0.90)
- CADS: Unleashing the Diversity of Diffusion Models through Condition-Annealed Sampling (ICLR 2024 · similarity 0.90)
- Unlocking Dataset Distillation with Diffusion Models (NeurIPS 2025 · similarity 0.90)
- Bridging Data Gaps in Diffusion Models with Adversarial Noise-Based Transfer Learning (ICML 2024 · similarity 0.89)
- Simpler Diffusion: 1.5 FID on ImageNet512 with Pixel-space Diffusion (CVPR 2025 · similarity 0.89)
- A Simple Early Exiting Framework for Accelerated Sampling in Diffusion Models (ICML 2024 · similarity 0.89)
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.