Research map / Image generation / Fast diffusion sampling
Efficient Diffusion Training via Min-SNR Weighting Strategy
Tiankai Hang, Shuyang Gu, Chen Li, Jianmin Bao, Dong Chen, Han Hu, Xin Geng, Baining Guo
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 |
|---|---|---|---|
| weight diffusion loss by min(SNR, gamma) with gamma=5 | ViT-B on ImageNet 256x256 · ImageNet 256x256 | convergence speed to FID 10: 3.4x speedup | Improved |
| add Min-SNR-5 weighting when predicting x0 | UNet · ImageNet 256x256, 200K iterations | FID: 25.93 -> 7.99 | Improved |
| add Min-SNR-5 weighting when predicting epsilon | UNet · ImageNet 256x256, 1M iterations | FID: 4.21 -> 4.14 | Improved |
| use plain SNR weighting instead of Min-SNR (ablation) | ViT-B diffusion training · ImageNet 256x256 | convergence | Hurt |
| use constant or Max-SNR-gamma weighting when network outputs noise (ablation) | diffusion training · ImageNet 256x256 | training stability | Hurt |
| increase gamma from 5 to 20 (ablation) | ViT-B predicting x0 · ImageNet 256x256 | FID: 4.92 -> 5.45 | Hurt |
| vary gamma on UNet predicting epsilon (ablation) | UNet · ImageNet 256x256 | FID: 4.14 -> 4.12 | No clear effect |
| use fixed Min-SNR-gamma weights instead of per-iteration Pareto/UGD optimized weights (ablation) | diffusion loss weighting · Pareto objective | Pareto objective value | No clear effect |
| apply Min-SNR-gamma to the EDM framework | EDM · ImageNet 64x64 | convergence | Improved |
Closest papers by meaning
- Denoising Task Difficulty-based Curriculum for Training Diffusion Models (ICLR 2025 · similarity 0.89)
- Dual-Rate Diffusion: Accelerating diffusion models with an interleaved heavy-light network (NeurIPS 2026 · similarity 0.89)
- Simpler Diffusion: 1.5 FID on ImageNet512 with Pixel-space Diffusion (CVPR 2025 · similarity 0.88)
- Decouple-Then-Merge: Finetune Diffusion Models as Multi-Task Learning (CVPR 2025 · similarity 0.88)
- Adaptive Non-Uniform Timestep Sampling for Accelerating Diffusion Model Training (CVPR 2025 · similarity 0.88)
- EfficientDM: Efficient Quantization-Aware Fine-Tuning of Low-Bit Diffusion Models (ICLR 2024 · 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.
