Domain Generalization on ImageNet-C

Metric: mean Corruption Error (mCE) (lower is better)

LeaderboardDataset

Loading chart...

Results

Hide extra data

Sort:

#	Model↕	mean Corruption Error (mCE)▲	Extra Data	Paper	Date↕	Code
1	DINOv2 (ViT-g/14, frozen model, linear eval)	28.2	Yes	DINOv2: Learning Robust Visual Features without ...	2023-04-14	Code
2	CAFormer-B36 (IN21K, 384)	30.8	Yes	MetaFormer Baselines for Vision	2022-10-24	Code
3	MAE+DAT (ViT-H)	31.4	No	Enhance the Visual Representation via Discrete A...	2022-09-16	Code
4	DINOv2 (ViT-L/14, frozen model, linear eval)	31.5	Yes	DINOv2: Learning Robust Visual Features without ...	2023-04-14	Code
5	CAFormer-B36 (IN21K)	31.8	Yes	MetaFormer Baselines for Vision	2022-10-24	Code
6	MAE (ViT-H)	33.8	No	Masked Autoencoders Are Scalable Vision Learners	2021-11-11	Code
7	ConvFormer-B36 (IN21K)	35	Yes	MetaFormer Baselines for Vision	2022-10-24	Code
8	FAN-L-Hybrid (IN-22k)	35.8	Yes	Understanding The Robustness in Vision Transform...	2022-04-26	Code
9	Pyramid Adversarial Training Improves ViT (Im21k)	36.8	Yes	Pyramid Adversarial Training Improves ViT Perfor...	2021-11-30	Code
10	VOLO-D5+HAT	38.4	No	Improving Vision Transformers by Revisiting High...	2022-04-03	Code
11	DiscreteViT (Im21k)	38.74	Yes	Discrete Representations Strengthen Vision Trans...	2021-11-20	Code
12	ConvNeXt-XL (Im21k) (augmentation overlap with ImageNet-C)	38.8	Yes	A ConvNet for the 2020s	2022-01-10	Code
13	GPaCo (ViT-L)	39	No	Generalized Parametric Contrastive Learning	2022-09-26	Code
14	FAN-B-Hybrid (IN-22k)	41	Yes	Understanding The Robustness in Vision Transform...	2022-04-26	Code
15	Pyramid Adversarial Training Improves ViT	41.42	No	Pyramid Adversarial Training Improves ViT Perfor...	2021-11-30	Code
16	FAN-L-Hybrid+STL	42.1	No	Fully Attentional Networks with Self-emerging To...	2024-01-08	Code
17	QualNet (ResNeXt101)	42.5	No	-	-	Code
18	CAFormer-B36	42.6	No	MetaFormer Baselines for Vision	2022-10-24	Code
19	DINOv2 (ViT-B/14, frozen model, linear eval)	42.7	Yes	DINOv2: Learning Robust Visual Features without ...	2023-04-14	Code
20	FAN-L-Hybrid	43	No	Understanding The Robustness in Vision Transform...	2022-04-26	Code
21	DrViT	46.22	No	Discrete Representations Strengthen Vision Trans...	2021-11-20	Code
22	DiscreteViT	46.22	No	Discrete Representations Strengthen Vision Trans...	2021-11-20	Code
23	ConvFormer-B36	46.3	No	MetaFormer Baselines for Vision	2022-10-24	Code
24	RVT-B*	46.8	No	Towards Robust Vision Transformer	2021-05-17	Code
25	Sequencer2D-L	48.9	No	Sequencer: Deep LSTM for Image Classification	2022-05-04	Code
26	RVT-S*	49.4	No	Towards Robust Vision Transformer	2021-05-17	Code
27	ResNet-50 (PushPull-Conv) + PRIME	49.95	No	PushPull-Net: Inhibition-driven ResNet robust to...	2024-08-07	Code
28	QualNet (ResNet-50)	50.6	No	-	-	Code
29	PRIME + DeepAugment (ResNet-50)	51.3	No	PRIME: A few primitives can boost robustness to ...	2021-12-27	Code
30	GFNet-S	53.8	No	Global Filter Networks for Image Classification	2021-07-01	Code
31	DINOv2 (ViT-S/14, frozen model, linear eval)	54.4	Yes	DINOv2: Learning Robust Visual Features without ...	2023-04-14	Code
32	PRIME with JSD (ResNet-50)	55.5	No	PRIME: A few primitives can boost robustness to ...	2021-12-27	Code
33	RVT-Ti*	57	No	Towards Robust Vision Transformer	2021-05-17	Code
34	PRIME (ResNet-50)	57.5	No	PRIME: A few primitives can boost robustness to ...	2021-12-27	Code
35	APR-SP + DeepAugment (ResNet-50)	57.5	No	Amplitude-Phase Recombination: Rethinking Robust...	2021-08-19	Code
36	DeepAugment (ResNet-50)	60.4	No	The Many Faces of Robustness: A Critical Analysi...	2020-06-29	Code
37	APR-SP (ResNet-50)	65	No	Amplitude-Phase Recombination: Rethinking Robust...	2021-08-19	Code
38	AugMix (ResNet-50)	65.3	No	AugMix: A Simple Data Processing Method to Impro...	2019-12-05	Code
39	Stylized ImageNet (ResNet-50)	69.3	Yes	ImageNet-trained CNNs are biased towards texture...	2018-11-29	Code
40	Group-wise Inhibition (ResNet-50)	69.6	No	Group-wise Inhibition based Feature Regularizati...	2021-03-03	Code
41	ResNet-50	76.7	No	Benchmarking Neural Network Robustness to Common...	2019-03-28	Code

#1DINOv2 (ViT-g/14, frozen model, linear eval)SOTA
28.2
mean Corruption Error (mCE)· Extra Data· 2023-04-14
DINOv2: Learning Robust Visual Features without Supervision Code
#2CAFormer-B36 (IN21K, 384)SOTA
30.8
mean Corruption Error (mCE)· Extra Data· 2022-10-24
MetaFormer Baselines for Vision Code
#3MAE+DAT (ViT-H)SOTA
31.4
mean Corruption Error (mCE)· 2022-09-16
Enhance the Visual Representation via Discrete Adversarial Training Code
#4DINOv2 (ViT-L/14, frozen model, linear eval)
31.5
mean Corruption Error (mCE)· Extra Data· 2023-04-14
DINOv2: Learning Robust Visual Features without Supervision Code
#5CAFormer-B36 (IN21K)
31.8
mean Corruption Error (mCE)· Extra Data· 2022-10-24
MetaFormer Baselines for Vision Code
#6MAE (ViT-H)SOTA
33.8
mean Corruption Error (mCE)· 2021-11-11
Masked Autoencoders Are Scalable Vision Learners Code
#7ConvFormer-B36 (IN21K)
35
mean Corruption Error (mCE)· Extra Data· 2022-10-24
MetaFormer Baselines for Vision Code
#8FAN-L-Hybrid (IN-22k)
35.8
mean Corruption Error (mCE)· Extra Data· 2022-04-26
Understanding The Robustness in Vision Transformers Code
#9Pyramid Adversarial Training Improves ViT (Im21k)
36.8
mean Corruption Error (mCE)· Extra Data· 2021-11-30
Pyramid Adversarial Training Improves ViT Performance Code
#10VOLO-D5+HAT
38.4
mean Corruption Error (mCE)· 2022-04-03
Improving Vision Transformers by Revisiting High-frequency Components Code
#11DiscreteViT (Im21k)
38.74
mean Corruption Error (mCE)· Extra Data· 2021-11-20
Discrete Representations Strengthen Vision Transformer Robustness Code
#12ConvNeXt-XL (Im21k) (augmentation overlap with ImageNet-C)
38.8
mean Corruption Error (mCE)· Extra Data· 2022-01-10
A ConvNet for the 2020s Code
#13GPaCo (ViT-L)
39
mean Corruption Error (mCE)· 2022-09-26
Generalized Parametric Contrastive Learning Code
#14FAN-B-Hybrid (IN-22k)
41
mean Corruption Error (mCE)· Extra Data· 2022-04-26
Understanding The Robustness in Vision Transformers Code
#15Pyramid Adversarial Training Improves ViT
41.42
mean Corruption Error (mCE)· 2021-11-30
Pyramid Adversarial Training Improves ViT Performance Code
#16FAN-L-Hybrid+STL
42.1
mean Corruption Error (mCE)· 2024-01-08
Fully Attentional Networks with Self-emerging Token Labeling Code
#17QualNet (ResNeXt101)
42.5
mean Corruption Error (mCE)
No paperCode
#18CAFormer-B36
42.6
mean Corruption Error (mCE)· 2022-10-24
MetaFormer Baselines for Vision Code
#19DINOv2 (ViT-B/14, frozen model, linear eval)
42.7
mean Corruption Error (mCE)· Extra Data· 2023-04-14
DINOv2: Learning Robust Visual Features without Supervision Code
#20FAN-L-Hybrid
43
mean Corruption Error (mCE)· 2022-04-26
Understanding The Robustness in Vision Transformers Code
#21DrViT
46.22
mean Corruption Error (mCE)· 2021-11-20
Discrete Representations Strengthen Vision Transformer Robustness Code
#22DiscreteViT
46.22
mean Corruption Error (mCE)· 2021-11-20
Discrete Representations Strengthen Vision Transformer Robustness Code
#23ConvFormer-B36
46.3
mean Corruption Error (mCE)· 2022-10-24
MetaFormer Baselines for Vision Code
#24RVT-B*SOTA
46.8
mean Corruption Error (mCE)· 2021-05-17
Towards Robust Vision Transformer Code
#25Sequencer2D-L
48.9
mean Corruption Error (mCE)· 2022-05-04
Sequencer: Deep LSTM for Image Classification Code
#26RVT-S*SOTA
49.4
mean Corruption Error (mCE)· 2021-05-17
Towards Robust Vision Transformer Code
#27ResNet-50 (PushPull-Conv) + PRIME
49.95
mean Corruption Error (mCE)· 2024-08-07
PushPull-Net: Inhibition-driven ResNet robust to image corruptions Code
#28QualNet (ResNet-50)
50.6
mean Corruption Error (mCE)
No paperCode
#29PRIME + DeepAugment (ResNet-50)
51.3
mean Corruption Error (mCE)· 2021-12-27
PRIME: A few primitives can boost robustness to common corruptions Code
#30GFNet-S
53.8
mean Corruption Error (mCE)· 2021-07-01
Global Filter Networks for Image Classification Code
#31DINOv2 (ViT-S/14, frozen model, linear eval)
54.4
mean Corruption Error (mCE)· Extra Data· 2023-04-14
DINOv2: Learning Robust Visual Features without Supervision Code
#32PRIME with JSD (ResNet-50)
55.5
mean Corruption Error (mCE)· 2021-12-27
PRIME: A few primitives can boost robustness to common corruptions Code
#33RVT-Ti*SOTA
57
mean Corruption Error (mCE)· 2021-05-17
Towards Robust Vision Transformer Code
#34PRIME (ResNet-50)
57.5
mean Corruption Error (mCE)· 2021-12-27
PRIME: A few primitives can boost robustness to common corruptions Code
#35APR-SP + DeepAugment (ResNet-50)
57.5
mean Corruption Error (mCE)· 2021-08-19
Amplitude-Phase Recombination: Rethinking Robustness of Convolutional Neural Networks in Frequency Domain Code
#36DeepAugment (ResNet-50)SOTA
60.4
mean Corruption Error (mCE)· 2020-06-29
The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization Code
#37APR-SP (ResNet-50)
65
mean Corruption Error (mCE)· 2021-08-19
Amplitude-Phase Recombination: Rethinking Robustness of Convolutional Neural Networks in Frequency Domain Code
#38AugMix (ResNet-50)SOTA
65.3
mean Corruption Error (mCE)· 2019-12-05
AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty Code
#39Stylized ImageNet (ResNet-50)SOTA
69.3
mean Corruption Error (mCE)· Extra Data· 2018-11-29
ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness Code
#40Group-wise Inhibition (ResNet-50)
69.6
mean Corruption Error (mCE)· 2021-03-03
Group-wise Inhibition based Feature Regularization for Robust Classification Code
#41ResNet-50
76.7
mean Corruption Error (mCE)· 2019-03-28
Benchmarking Neural Network Robustness to Common Corruptions and Perturbations Code