Research map / Image generation / Fast diffusion sampling
Beta-Tuned Timestep Diffusion Model
Tianyi Zheng, Peng-Tao Jiang, Ben Wan, Hao Zhang, Jinwei Chen, Jia Wang, Bo Li
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 uniform timestep sampling with Beta(alpha<1, beta=1) sampling | DDPM (ADM architecture) · FFHQ 128x128 | FID: 14.07 -> 10.28 | Improved |
| replace uniform timestep sampling with beta sampling | DDPM · ImageNet 32x32 | FID: 3.65 -> 2.87 | Improved |
| add beta timestep sampling | DDPM · ImageNet 32x32 training speed | training speedup to equal FID: 4.71x | Improved |
| add beta timestep sampling | DDPM · AFHQ-D 256x256 training speed | training speedup to equal FID: 3.33x | Improved |
| use loss re-weighting methods (Debias, Min-SNR, P2-Weight) on low-resolution CIFAR | DDPM · CIFAR-10 32x32 | FID: 3.01 -> 3.17 / 3.23 / 3.19 | Hurt |
| add B-TTDM to P2-Weight | P2-Weight · AFHQ-D 256x256 | FID: 16.18 -> 12.45 | Improved |
| add B-TTDM to Min-SNR | Min-SNR · AFHQ-D 256x256 | FID: 17.03 -> 13.34 | Improved |
| add B-TTDM to DDPM-IP | DDPM-IP · CIFAR-10 32x32 | FID: 2.72 -> 2.46 | Improved |
| use B-TTDM with fast samplers (DDIM, DPM-Solver) at 25 NFE | DDPM · AFHQ-D 256x256 | FID: 24.22 -> 21.76 (DDIM, 25 NFE) | Improved |
| use alpha=0.8, beta=1 instead of uniform alpha=1, beta=1 (ablation) | B-TTDM · AFHQ-D 256x256 | FID: 17.21 -> 13.81 | Improved |
| lower alpha from 0.8 to 0.7 (ablation) | B-TTDM · AFHQ-D 256x256 | FID: 13.81 -> 14.78 | Hurt |
| use beta sampling instead of exponential timestep sampling (ablation) | B-TTDM · AFHQ 256x256 | FID: 15.12 -> 13.81 | Improved |
Closest papers by meaning
- Adaptive Non-Uniform Timestep Sampling for Accelerating Diffusion Model Training (CVPR 2025 · similarity 0.91)
- Towards More Accurate Diffusion Model Acceleration with A Timestep Tuner (CVPR 2024 · similarity 0.88)
- Diffusion Model is Effectively Its Own Teacher (CVPR 2025 · similarity 0.87)
- Simple and Fast Distillation of Diffusion Models (NeurIPS 2024 · similarity 0.87)
- Distribution Backtracking Builds A Faster Convergence Trajectory for Diffusion Distillation (ICLR 2025 · similarity 0.87)
- Tuning Timestep-Distilled Diffusion Model Using Pairwise Sample Optimization (ICLR 2025 · 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.
