AI Paper Atlas

Research map / Benchmarks / ImageNet-C

ImageNet-C leaderboard

93 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on ImageNet-C. Best mCE: 13.6 (SATA-B*, CVPR 2025).

Numbers are as each paper reports them, read by AI from its results tables and not checked by hand. Papers can differ in setting (training data, model size, extra data); the topic column shows what each paper is about. Open the paper before you compare or quote a number.

ImageNet-C · mCE

68 papers, each with its best reported number for its own method; lower is better.

#mCE ↓MethodPaperTopic
113.6SATA-B*
ViT-Base/16
SATA: Spatial Autocorrelation Token Analysis for Enhancing the Robustness of Vision Transformers
CVPR 2025 · Table 1
Attention and transformers
229.5ELEAS
ViT-L
Edges to Shapes to Concepts: Adversarial Augmentation for Robust Vision
CVPR 2023 · Table 1
Attacks on text and graphs
334.4EdgeNet-ViT-B/16
ViT-B/16
Harnessing Edge Information for Improved Robustness in Vision Transformers
AAAI 2024 · Table 2
Certified and empirical robustness
434.4TORA-ViT-B/16 (λ=0.5)
ViT-B/16
Trade-Off Between Robustness and Accuracy of Vision Transformers
CVPR 2023 · Table 1
Certified and empirical robustness
535.8FAN-L-HybridUnderstanding The Robustness in Vision Transformers
ICML 2022 · Table: Main results on out-of-
CNN features
636.5RepLKNet-31BRevealing the Dark Secrets of Extremely Large Kernel ConvNets on Robustness
ICML 2024 · Table: mCEs on ImageNet-C
CNN design
738DTN
ViT-S*
Dynamic Token Normalization improves Vision Transformers
ICLR 2022 · Appendix ImageNet-C table
Attention and transformers
838DeepAug+AM+RNXt101 + BN adapt
ResNeXt-101
Improving robustness against common corruptions by covariate shift adaptation
NeurIPS 2020 · Table 1
Test-time adaptation
938.2XTRA
ViT-B
The Extra Tokens Matter: Disentangled Representation Learning with Vision Transformers
ICML 2026 · Table 4
Attention and transformers
1039.8GFN (FAN)
FAN
Gain from Neighbors: Boosting Model Robustness in the Wild via Adversarial Perturbations Toward Neighboring Classes
CVPR 2025 · Table 1
Certified and empirical robustness
58 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ImageNet-C · accuracy

30 papers, each with its best reported number for its own method; higher is better.

#accuracy ↑MethodPaperTopic
159.9LookupViT 10x10
B/16
LookupViT: Compressing visual information to a limited number of tokens
ECCV 2024 · Table 2
Attention and transformers
258.8AugMix + P-Rotate / L2
ViT-B/16 + AugMix
Understanding and Improving Robustness of Vision Transformers through Patch-based Negative Augmentation
NeurIPS 2022 · Table 2
Recognition and long-tail
356.8R-SAM
ViT-B-16
Random Sharpness-Aware Minimization
NeurIPS 2022 · Table 4
Implicit bias and generalization
456.5ViT-B/16-SAM
ViT-B/16
When Vision Transformers Outperform ResNets without Pre-training or Strong Data Augmentations
ICLR 2022 · Table: Performance of ResNets,
Attention and transformers
556.1TRUST
VMamba
TRUST: Test-Time Refinement using Uncertainty-Guided SSM Traverses
NeurIPS 2025 · Table: Top-1 classification ac
Test-time adaptation
655DINO + SDMP
ViT-S
A Simple Data Mixing Prior for Improving Self-Supervised Learning
CVPR 2022 · Table 4
Self-supervised and contrastive
751.87CR-SAM
ResNet-101
CR-SAM: Curvature Regularized Sharpness-Aware Minimization
AAAI 2024 · ImageNet results table
Implicit bias and generalization
851.7GSAM
ViT-B/32
Surrogate Gap Minimization Improves Sharpness-Aware Training
ICLR 2022 · Table 1
Implicit bias and generalization
951.2Auto-RGNSurgical Fine-Tuning Improves Adaptation to Distribution Shifts
ICLR 2023 · Table: Results on Real World D
Recognition and long-tail
1050.7Vanilla ResNet-50 + BN adapt
ResNet-50
Improving robustness against common corruptions by covariate shift adaptation
NeurIPS 2020 · Table 1
Test-time adaptation
20 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

Not affiliated with any conference, journal or the benchmark's authors. Corrections: contact@latenticlabs.com.