Research map / Image generation / Fast diffusion sampling
OBS-Diff: Accurate Pruning For Diffusion Models in One-Shot
Junhan Zhu, Hesong Wang, Mingluo Su, Zefang Wang, Huan Wang
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 |
|---|---|---|---|
| use OBS-Diff instead of Wanda at 60% unstructured sparsity | SD 3-medium · MS-COCO 2014 | FID: 170.33 -> 28.49 | Improved |
| use OBS-Diff instead of Wanda at 70% unstructured sparsity | Flux 1.dev · MS-COCO 2014 | FID: 49.68 -> 39.79 | Improved |
| use OBS-Diff instead of Wanda at 50% unstructured sparsity | SD 3.5-large · MS-COCO 2014 | FID: 27.49 -> 29.61 | Hurt |
| use OBS-Diff instead of Wanda at 2:4 sparsity | SD 3.5-large · MS-COCO 2014 | FID: 32.08 -> 32.13 | No clear effect |
| use OBS-Diff instead of Wanda at 2:4 sparsity | SD 3.5-large · MS-COCO 2014 | ImageReward: -0.1363 -> 0.4493 | Improved |
| use OBS-Diff instead of L1-norm at 30% structured sparsity | SD 3.5-large · MS-COCO 2014 | FID: 327.48 -> 34.51 | Improved |
| use OBS-Diff instead of EcoDiff at 30% structured sparsity | SDXL · MS-COCO 2014 | FID: 101.96 -> 29.75 | Improved |
| use OBS-Diff instead of Diff-Pruning | DDPM CIFAR-10 · CIFAR-10 | FID: 7.72 -> 7.55 | Improved |
| add timestep-aware Hessian weighting with logarithmic decrease (ablation) | OBS-Diff · MS-COCO 2014 (1,000 prompts) | ImageReward | Improved |
| increase the number of module packages (ablation) | OBS-Diff · MS-COCO 2014 (1,000 prompts) | ImageReward | No clear effect |
| increase calibration prompts up to 100 (ablation) | OBS-Diff · MS-COCO 2014 (1,000 prompts) | ImageReward | Improved |
| calibrate on out-of-distribution GCC3M instead of MS-COCO train (ablation) | OBS-Diff SD 3-medium 50% unstructured · MS-COCO 2014 | FID: 27.93 -> 27.20 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports OBS-Diff as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Importance-Aware OBS Pruning for Diffusion Models (NeurIPS 2026 · better on 14 of 14 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
- DiP-GO: A Diffusion Pruner via Few-step Gradient Optimization (NeurIPS 2024 · similarity 0.91)
- ALTER: All-in-One Layer Pruning and Temporal Expert Routing for Efficient Diffusion Generation (NeurIPS 2025 · similarity 0.91)
- Attention-Driven Training-Free Efficiency Enhancement of Diffusion Models (CVPR 2024 · similarity 0.89)
- Simpler Diffusion: 1.5 FID on ImageNet512 with Pixel-space Diffusion (CVPR 2025 · similarity 0.89)
- Neural Diffusion Models (ICML 2024 · similarity 0.89)
- Diffusion Models Without Attention (CVPR 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.
