AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

A Comprehensive Study on Robustness of Image Classification Models: Benchmarking and Rethinking

Chang Liu, Yinpeng Dong, Wenzhao Xiang, Xiao Hui Yang, Hang Su, Jun Zhu, Yuefeng Chen, Yuan He, Hui Xue, Shibao Zheng

IJCV 2024 · Label-efficient and robust learning › Recognition and long-tail · approach: Learning from unlabeled and positive-unlabeled data · 163 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 RandAugment (ablation)SwinS AT · ImageNet AutoAttack eps=4/255robust accuracy: 48.7 -> 49.7Improved
add Mixup on top of RandAugment (ablation)SwinS AT · ImageNet AutoAttack eps=4/255robust accuracy: 49.7 -> 54.9Improved
add label smoothing 0.1 (ablation)SwinS AT · ImageNet AutoAttack eps=4/255robust accuracy: 54.9 -> 55.4Improved
increase weight decay from 0 to 0.1 (ablation)SwinS AT · ImageNet AutoAttack eps=4/255robust accuracy: 55.4 -> 53.7Hurt
add EMA weight averaging (ablation)SwinS AT · ImageNet AutoAttack eps=4/255robust accuracy: 55.4 -> 55.6Improved
initialize AT from ImageNet-21K pre-training (ablation)SwinB AT · ImageNet AutoAttack eps=4/255robust accuracy: 57.3 -> 57.4No clear effect
initialize AT from SimMIM (ablation)SwinB AT · ImageNet AutoAttack eps=4/255robust accuracy: 57.3 -> 55.9No clear effect
initialize AT from CLIP image encoder (ablation)ViTB AT · ImageNet AutoAttack eps=4/255robust accuracy: 49.7 -> 42.8Hurt
apply adversarial trainingViTL/ResNet family · natural robustness average over 10 datasetsaverage accuracy: 34.1 -> 30.2Hurt
apply adversarial trainingViTB · natural robustness average over 10 datasetsaverage accuracy: 35.8 -> 39.1Improved
add ImageNet-21K pre-trainingViTL · natural robustness average over 10 datasetsaverage accuracy: 35.4 -> 57.4Improved
add MAE self-supervised pre-trainingViTL · natural robustness average over 10 datasetsaverage accuracy: 35.4 -> 53.6Improved

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