Research map / Architectures and efficiency / RNNs and normalization
Squeeze, Recover and Relabel: Dataset Condensation at ImageNet Scale From A New Perspective
Zeyuan Yin, Eric Xing, Zhiqiang Shen
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 multi-crop optimization during recovery (ablation) | SRe2L · ImageNet-1K | top-1 accuracy: 40.37 -> 46.71 | Improved |
| add multi-crop optimization during recovery (ablation) | SRe2L · Tiny-ImageNet | top-1 accuracy: 30.30 -> 37.88 | Improved |
| add TV and L2 image prior regularizers (ablation) | SRe2L recovery · ImageNet-1K | top-1 accuracy: 40.37 -> 22.92 | Hurt |
| add TV and L2 image prior regularizers (ablation) | SRe2L recovery · Tiny-ImageNet | top-1 accuracy: 30.30 -> 29.87 | No clear effect |
| increase squeeze budget from 50 to 400 iterations (ablation) | SRe2L squeeze stage · Tiny-ImageNet | top-1 accuracy: 37.88 -> 23.88 | Hurt |
| add Mixup or CutMix augmentation in squeeze stage (ablation) | SRe2L squeeze stage · Tiny-ImageNet | top-1 accuracy: 37.88 -> 35.59 (Mixup), 25.90 (CutMix) | Hurt |
| increase recovery budget from 0.5k to 4k iterations (ablation) | SRe2L recovery · ImageNet-1K | top-1 accuracy: 38.94 -> 46.73 | Improved |
| increase recovery budget from 2k to 4k iterations (ablation) | SRe2L recovery ResNet-18 · Tiny-ImageNet | top-1 accuracy: 41.35 -> 41.11 | No clear effect |
| increase image resolution from 64x64 to 224x224 (ablation) | SRe2L · ImageNet-1K IPC 50 | top-1 accuracy: 35.27 -> 46.75 | Improved |
| use larger student/recovery architecture R18 to R101 | SRe2L · ImageNet-1K IPC 50 | top-1 accuracy: 46.8 -> 60.8 | Improved |
| use larger architecture R18 to R101 | SRe2L · Tiny-ImageNet IPC 50 | top-1 accuracy: 41.1 -> 42.5 | No clear effect |
Closest papers by meaning
- Elucidating the Design Space of Dataset Condensation (NeurIPS 2024 · similarity 0.90)
- Are Large-scale Soft Labels Necessary for Large-scale Dataset Distillation? (NeurIPS 2024 · similarity 0.89)
- Multisize Dataset Condensation (ICLR 2024 · similarity 0.89)
- M3D: Dataset Condensation by Minimizing Maximum Mean Discrepancy (AAAI 2024 · similarity 0.87)
- Rethinking Dataset Distillation for Classification: Do Distilled Sets Outperform Coresets? (NeurIPS 2026 · similarity 0.86)
- Unlocking Dataset Distillation with Diffusion Models (NeurIPS 2025 · similarity 0.86)
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.
