AI Paper Atlas

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

NeurIPS 2023 · Architectures and efficiency › RNNs and normalization · approach: Dataset condensation and flow estimation · 11 citations (OpenAlex)

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 changedWhereMeasuredEffect
add multi-crop optimization during recovery (ablation)SRe2L · ImageNet-1Ktop-1 accuracy: 40.37 -> 46.71Improved
add multi-crop optimization during recovery (ablation)SRe2L · Tiny-ImageNettop-1 accuracy: 30.30 -> 37.88Improved
add TV and L2 image prior regularizers (ablation)SRe2L recovery · ImageNet-1Ktop-1 accuracy: 40.37 -> 22.92Hurt
add TV and L2 image prior regularizers (ablation)SRe2L recovery · Tiny-ImageNettop-1 accuracy: 30.30 -> 29.87No clear effect
increase squeeze budget from 50 to 400 iterations (ablation)SRe2L squeeze stage · Tiny-ImageNettop-1 accuracy: 37.88 -> 23.88Hurt
add Mixup or CutMix augmentation in squeeze stage (ablation)SRe2L squeeze stage · Tiny-ImageNettop-1 accuracy: 37.88 -> 35.59 (Mixup), 25.90 (CutMix)Hurt
increase recovery budget from 0.5k to 4k iterations (ablation)SRe2L recovery · ImageNet-1Ktop-1 accuracy: 38.94 -> 46.73Improved
increase recovery budget from 2k to 4k iterations (ablation)SRe2L recovery ResNet-18 · Tiny-ImageNettop-1 accuracy: 41.35 -> 41.11No clear effect
increase image resolution from 64x64 to 224x224 (ablation)SRe2L · ImageNet-1K IPC 50top-1 accuracy: 35.27 -> 46.75Improved
use larger student/recovery architecture R18 to R101SRe2L · ImageNet-1K IPC 50top-1 accuracy: 46.8 -> 60.8Improved
use larger architecture R18 to R101SRe2L · Tiny-ImageNet IPC 50top-1 accuracy: 41.1 -> 42.5No clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy