Research map / Image generation / Fast diffusion sampling
Towards Consistent and Efficient Dataset Distillation via Diffusion-Driven Selection
Xinhao Zhong, Shuoyang Sun, Zhaoyang Xu, Xulin Gu, Bin Chen, Min Zhang, Yaowei 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 |
|---|---|---|---|
| add DIFT-feature clustering of patches (ablation) | diffusion patch selection · ImageWoof IPC=10 | accuracy: 41.8 -> 44.5 | Improved |
| add DIFT-feature clustering of patches (ablation) | diffusion patch selection · ImageNet-1K IPC=10 | accuracy: 43.6 -> 46.1 | Improved |
| add DIFT-feature clustering of patches (ablation) | diffusion patch selection · ImageNet-1K IPC=50 | accuracy: 58.1 -> 61.0 | Improved |
| add DIFT-feature clustering of patches (ablation) | diffusion patch selection · ImageNet-1K IPC=100 | accuracy: 62.0 -> 63.4 | Improved |
| cluster with DIFT instead of CLIP features (ablation) | diffusion patch selection · ImageWoof IPC=10 | accuracy: 39.8 -> 44.5 | Improved |
| cluster with DIFT instead of CLIP features (ablation) | diffusion patch selection · ImageNet-1K IPC=10 | accuracy: 42.1 -> 46.1 | Improved |
| cluster with CLIP features instead of no clustering (ablation) | diffusion patch selection · ImageWoof IPC=10 | accuracy: 41.8 -> 39.8 | Hurt |
| use time-step range 0.1-0.7 instead of 0.7-0.9 (ablation) | noise-loss difference scoring · ImageNet-1K IPC=50 | accuracy: 57.9 -> 61.0 | Improved |
| use 224x224 patches instead of 112x112 from 256x256 images (ablation) | patch cropping · ImageWoof IPC=50 | accuracy: 67.4 -> 73.7 | Improved |
| use 448x448 patches instead of 224x224 from 512x512 images (ablation) | patch cropping · ImageWoof IPC=100 | accuracy: 78.4 -> 81.0 | Improved |
| use diffusion patch selection instead of RDED | ImageNet-1K distillation · ImageNet-1K IPC=10 | accuracy: 42.0 -> 46.1 | Improved |
| use diffusion patch selection instead of D4M | ImageNet-1K distillation · ImageNet-1K IPC=10 | accuracy: 27.9 -> 46.1 | Improved |
Closest papers by meaning
- Unlocking Dataset Distillation with Diffusion Models (NeurIPS 2025 · similarity 0.94)
- Efficient Dataset Distillation via Minimax Diffusion (CVPR 2024 · similarity 0.93)
- Taming Diffusion for Dataset Distillation with High Representativeness (ICML 2025 · similarity 0.93)
- MGD$^3$ : Mode-Guided Dataset Distillation using Diffusion Models (ICML 2025 · similarity 0.93)
- D^4: Dataset Distillation via Disentangled Diffusion Model (CVPR 2024 · similarity 0.92)
- Towards Universal Dataset Distillation via Task-Driven Diffusion (CVPR 2025 · 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.
