Image Classification on ImageNet-C (test)
100Defocus Blur AccResNet50
Evaluation results
| Method | Links | ||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ResNet50Backbone=ResNet502024.08 | 100 | — | 1 | 1 | 100 | 1 | 1 | 1 | 100 | 1 | 1 | 1 | 100 | 100 | 1 | 66.7 | — | 1 | — | — | 1 | 26.9 | — | — | |
| Res50+AugMixBackbone=ResNet50, Data Augmentation=AugMix2024.08 | 100 | — | 1 | 1 | 100 | 1 | 1 | 1 | 100 | 1 | 1 | 1 | 100 | 100 | 1 | 61.21 | — | 1 | — | — | 1 | 26.9 | — | — | |
| Res50+AutoAugBackbone=ResNet50, Data Augmentation=AutoAug2024.08 | 100 | — | 1 | 1 | 100 | 1 | 1 | 1 | 100 | 1 | 1 | 1 | 100 | 100 | 1 | 62.98 | — | 1 | — | — | 1 | 26.9 | — | — | |
| Res50+PRIMEBackbone=ResNet50, Data Augmentation=PRIME2024.08 | 100 | — | 1 | 1 | 100 | 1 | 1 | 1 | 100 | 1 | 1 | 1 | 100 | 100 | 1 | 52.23 | — | 1 | — | — | 1 | 29.8 | — | — | |
| Res50+TrivialAugBackbone=ResNet50, Data Augmentation=TrivialAug2024.08 | 100 | — | 1 | 1 | 100 | 1 | 1 | 1 | 100 | 1 | 1 | 1 | 100 | 100 | 1 | 59.61 | — | 1 | — | — | 1 | 26.9 | — | — | |
| Res50+TrivialAug+PP3Backbone=ResNet50, Data Augmentation=TrivialAug, Inhibition Mechanism=PP32024.08 | 100 | — | 1.02 | 0.99 | 96 | 1.01 | 0.99 | 1 | 100 | 1.07 | 1.03 | 0.96 | 92 | 83 | 0.97 | 58.67 | — | 0.985 | — | — | 1.02 | 28.6 | — | — | |
| EffNet-b0Backbone=EfficientNet-b02024.08 | 100 | — | 1 | 1 | 100 | 1 | 1 | 1 | 100 | 1 | 1 | 1 | 100 | 100 | 1 | 68.18 | — | 1 | — | — | 1 | 28.6 | — | — | |
| PushPullBackbone=ResNet50, Inhibition Mechanism=[32]2024.08 | 99 | — | 1 | 1.02 | 100 | 0.99 | 0.99 | 0.99 | 98 | 1 | 0.98 | 0.99 | 99 | 99 | 1.03 | 66.5 | — | 0.996 | — | — | 1 | 26.9 | — | — | |
| BlurConv2Backbone=ResNet50, Inhibition Mechanism=BlurConv22024.08 | 99 | — | 0.89 | 0.89 | 97 | 0.99 | 0.96 | 1 | 96 | 0.98 | 1 | 1 | 98 | 70 | 0.87 | 62.1 | — | 0.938 | — | — | 0.9 | 27.6 | — | — | |
| Res50+PRIME+PP3Backbone=ResNet50, Data Augmentation=PRIME, Inhibition Mechanism=PP32024.08 | 99 | — | 0.9 | 0.89 | 97 | 1 | 1.01 | 0.99 | 98 | 1 | 1 | 0.96 | 96 | 84 | 0.9 | 49.95 | — | 0.954 | — | — | 0.91 | 30.6 | — | — | |
| EffNet-b0 + PP3Backbone=EfficientNet-b0, Inhibition Mechanism=PP32024.08 | 99 | — | 0.99 | 0.97 | 98 | 1 | 1 | 1 | 99 | 1 | 0.99 | 1.01 | 98 | 91 | 0.86 | 66.6 | — | 0.976 | — | — | 0.98 | 28.8 | — | — | |
| PushPull avg5 (PP5)Backbone=ResNet50, Inhibition Mechanism=PP52024.08 | 98 | — | 0.98 | 0.99 | 95 | 0.97 | 0.97 | 0.97 | 97 | 0.99 | 0.98 | 0.99 | 92 | 83 | 1 | 64.5 | — | 0.966 | — | — | 0.99 | 27.6 | — | — | |
| Res50+AugMix+PP3Backbone=ResNet50, Data Augmentation=AugMix, Inhibition Mechanism=PP32024.08 | 98 | — | 0.94 | 0.93 | 94 | 0.99 | 0.95 | 1.01 | 99 | 1.06 | 1.02 | 1 | 89 | 84 | 1.02 | 59 | — | 0.968 | — | — | 0.95 | 28.7 | — | — | |
| BlurConv1Backbone=ResNet50, Inhibition Mechanism=BlurConv12024.08 | 97 | — | 0.96 | 0.97 | 97 | 0.98 | 0.97 | 0.97 | 96 | 0.95 | 0.96 | 0.97 | 95 | 94 | 0.95 | 62.6 | — | 0.962 | — | — | 0.97 | 26.5 | — | — | |
| Res50+AutoAug+PP3Backbone=ResNet50, Data Augmentation=AutoAug, Inhibition Mechanism=PP32024.08 | 97 | — | 0.95 | 0.94 | 97 | 0.99 | 0.98 | 1 | 98 | 1.02 | 1.01 | 0.95 | 93 | 82 | 0.96 | 60.41 | — | 0.961 | — | — | 0.95 | 28.3 | — | — | |
| PushPull avg3 (PP3)Backbone=ResNet50, Inhibition Mechanism=PP32024.08 | 96 | — | 0.98 | 0.98 | 96 | 0.98 | 0.97 | 0.99 | 100 | 1.02 | 0.99 | 0.99 | 91 | 81 | 0.97 | 64.5 | — | 0.967 | — | — | 0.99 | 28.2 | — | — | |
| BlurConv2+PP3Backbone=ResNet50, Inhibition Mechanism=BlurConv2 + PP32024.08 | 95 | — | 0.89 | 0.89 | 89 | 0.96 | 0.92 | 1.01 | 97 | 1.09 | 1.08 | 1.03 | 91 | 55 | 0.87 | 61.66 | — | 0.924 | — | — | 0.91 | 30.7 | — | — | |
| BN statsBackbone=ResNet-502026.07 | 85.2 | 85 | 84.2 | 85 | 84.4 | 73.2 | 61.2 | 65.9 | 68.1 | 51.9 | 34.8 | 83.1 | 55.9 | 51.3 | 59.9 | 68.6 | — | — | — | — | — | — | — | — | |
| T3ABackbone=ResNet-502026.07 | 83.7 | 98.4 | 98 | 98.9 | 91.7 | 87.4 | 80.4 | 86 | 79.6 | 79.4 | 41.9 | 95.7 | 86 | 82.4 | 71.9 | 84.1 | — | — | — | — | — | — | — | — | |
| StandardBackbone=ResNet-502019.12 | 82 | 79 | 80 | 82 | 90 | 84 | 80 | 86 | 81 | 75 | 65 | 79 | 91 | 77 | 80 | 80.6 | — | — | — | — | — | — | — | — | |
| No AdaptBackbone=ResNet-502026.07 | 81.7 | 97.8 | 97.1 | 98.2 | 89.8 | 85.2 | 78 | 83.5 | 77.1 | 75.9 | 41.3 | 94.6 | 82.5 | 79.3 | 68.6 | 82 | — | — | — | — | — | — | — | — | |
| Random AA*Backbone=ResNet-502019.12 | 80 | 70 | 71 | 72 | 86 | 82 | 81 | 81 | 77 | 72 | 61 | 75 | 88 | 73 | 72 | 76.1 | — | — | — | — | — | — | — | — | |
| ResNet-50 + l∞ Adversarial TrainingBackbone=ResNet-50, Adversarial Training=l∞2020.06 | 80 | — | — | — | 71 | 72 | 71 | — | — | — | — | — | — | — | — | — | — | — | 74 | — | — | — | — | — | |
| AutoAugment* (AA)Backbone=ResNet-502019.12 | 77 | 69 | 68 | 72 | 83 | 80 | 81 | 79 | 75 | 64 | 56 | 70 | 88 | 57 | 71 | 72.7 | — | — | — | — | — | — | — | — | |
| SINBackbone=ResNet-502019.12 | 77 | 69 | 70 | 70 | 80 | 66 | 66 | 74 | 75 | 69 | 65 | 69 | 80 | 64 | 77 | 73.3 | — | — | — | — | — | — | — | — | |
| CoLA (ROID)Backbone=ResNet-50, Source Model Weights=ROID [44]2026.07 | 77 | 74.8 | 74.5 | 74.1 | 76 | 65 | 55.2 | 59.3 | 61.8 | 47.4 | 34.2 | 72.3 | 50.2 | 44.9 | 52.1 | 61.3 | — | — | — | — | — | — | — | — | |
| CoLA (DeYO)Backbone=ResNet-50, Source Model Weights=DeYO [40]2026.07 | 76.8 | 74.3 | 73.5 | 73.6 | 74.8 | 65.5 | 56 | 60.5 | 62.4 | 49.3 | 35.6 | 72.3 | 49.9 | 46 | 53 | 61.6 | — | — | — | — | — | — | — | — | |
| Ours (DeYO)Backbone=ResNet-50, Source Model Weights=DeYO [40]2026.07 | 76.6 | 72.4 | 71.3 | 71.9 | 74 | 66.1 | 56.6 | 61 | 62.5 | 50.7 | 37.2 | 72.1 | 49.9 | 46.3 | 52.6 | 61.4 | — | — | — | — | — | — | — | — | |
| Ours (ROID)Backbone=ResNet-50, Source Model Weights=ROID [44]2026.07 | 74.4 | 70.4 | 69.7 | 69.8 | 72.8 | 63.5 | 54.9 | 58.9 | 61.1 | 47.5 | 35.8 | 69.2 | 49 | 44.2 | 50.2 | 59.4 | — | — | — | — | — | — | — | — | |
| Patch UniformBackbone=ResNet-502019.12 | 74 | 67 | 68 | 70 | 83 | 81 | 77 | 80 | 74 | 75 | 62 | 77 | 84 | 71 | 71 | 74.3 | — | — | — | — | — | — | — | — | |
| MaxBlur poolBackbone=ResNet-502019.12 | 74 | 73 | 74 | 76 | 86 | 78 | 77 | 77 | 72 | 63 | 56 | 68 | 86 | 71 | 71 | 73.4 | — | — | — | — | — | — | — | — | |
| AUGMIXBackbone=ResNet-502019.12 | 70 | 65 | 66 | 67 | 80 | 66 | 66 | 75 | 72 | 67 | 58 | 58 | 79 | 69 | 69 | 68.4 | — | — | — | — | — | — | — | — | |
| AUGMIX+SINBackbone=ResNet-502019.12 | 69 | 61 | 62 | 61 | 77 | 63 | 72 | 66 | 68 | 63 | 59 | 52 | 74 | 60 | 67 | 64.9 | — | — | — | — | — | — | — | — | |
| MVT + FTVLM=MMICL, Fine-tuned=true, Severity level=12025.12 | 68.8 | 71 | 72 | 68.5 | 68.9 | 72.8 | 65.2 | 70.6 | 71.9 | 72.5 | 76 | 73.5 | 70.7 | 72.8 | 71.2 | — | — | — | — | — | — | — | — | — | |
| No AdaptBackbone=ViT B/162026.07 | 68.8 | 65.8 | 67.3 | 65.3 | 74.4 | 64.3 | 66.6 | 56.8 | 45.2 | 48.6 | 29.2 | 81.8 | 57.1 | 60.8 | 50.2 | 60.2 | — | — | — | — | — | — | — | — | |
| BN statsBackbone=ViT B/162026.07 | 68.8 | 65.8 | 67.3 | 65.3 | 74.4 | 64.3 | 66.6 | 56.8 | 45.2 | 48.6 | 29.2 | 81.8 | 57.1 | 60.8 | 50.2 | 60.2 | — | — | — | — | — | — | — | — | |
| T3ABackbone=ViT B/162026.07 | 68.8 | 65.8 | 67.4 | 65.3 | 74.2 | 64.4 | 66.3 | 56.7 | 45.4 | 48.6 | 29.2 | 89.5 | 56.9 | 60.8 | 50.1 | 60.6 | — | — | — | — | — | — | — | — | |
| Machine Vision Therapy (MVT)VLM=MMICL, Severity level=12025.12 | 67.1 | 70.1 | 70.8 | 66.3 | 67.1 | 71.9 | 64.1 | 69.2 | 69.5 | 70.7 | 74.7 | 70.9 | 70 | 71.7 | 72.5 | — | — | — | — | — | — | — | — | — | |
| ResNet-152Backbone=ResNet-152, Model Architecture=Larger Models2020.06 | 67 | — | — | — | 81 | 66 | 74 | — | — | — | — | — | — | — | — | — | — | — | 58 | — | — | — | — | — | |
| ResNet-152Backbone=ResNet-1522021.11 | 66.9 | 72.5 | 73.4 | 76.3 | 81.4 | 65.7 | 74.5 | 70.7 | 67.8 | 62.1 | 51 | 67.1 | 75.6 | 68.9 | 65.1 | 69.27 | — | — | — | — | — | — | — | — | |
| ResNet-152 + StylizedBackbone=ResNet-152, Training Strategy=Stylized Training2021.11 | 66.1 | 63.3 | 63.1 | 64.6 | 77 | 63.5 | 71.6 | 62.4 | 65.4 | 59.4 | 52 | 62 | 73.2 | 55.3 | 62.9 | 64.19 | — | — | — | — | — | — | — | — | |
| CLIPBackbone=ViT-L, Severity level=12025.12 | 66.1 | 69.8 | 70.5 | 65.7 | 65.5 | 70.9 | 62.2 | 68.3 | 68.5 | 69.8 | 74.3 | 70.6 | 69.2 | 71 | 70.8 | — | — | — | — | — | — | — | — | — | |
| CoLA (DeYO)Backbone=ViT B/16, Source Model Weights=DeYO [40]2026.07 | 64.7 | 62.1 | 62.8 | 61.5 | 69.9 | 60.2 | 62.6 | 54.6 | 45.6 | 45.2 | 28 | 73.5 | 54.8 | 56 | 47 | 56.6 | — | — | — | — | — | — | — | — | |
| CoLA (ROID)Backbone=ViT B/16, Source Model Weights=ROID [44]2026.07 | 64.6 | 63.6 | 64.8 | 62.2 | 70.7 | 61.4 | 64.6 | 55.3 | 44.8 | 50.7 | 27.9 | 78.4 | 54.2 | 56.2 | 47.1 | 57.8 | — | — | — | — | — | — | — | — | |
| Ours (DeYO)Backbone=ViT B/16, Source Model Weights=DeYO [40]2026.07 | 64.2 | 60.3 | 61.5 | 60.3 | 68.3 | 58.6 | 61.2 | 54.4 | 47.1 | 46.1 | 27.9 | 75.4 | 53.8 | 53.3 | 45.4 | 55.9 | — | — | — | — | — | — | — | — | |
| ViT-BBackbone=ViT-B2021.11 | 63.2 | 43 | 47.2 | 44.4 | 73.4 | 55.1 | 70.8 | 51.6 | 45.6 | 35.2 | 44.2 | 41.3 | 61.6 | 54 | 59.4 | 52.71 | — | — | — | — | — | — | — | — | |
| ResNet-50Backbone=ResNet-502020.06 | 61 | — | — | — | 73 | 61 | 64 | — | — | — | — | — | — | — | — | — | — | — | 65 | — | — | — | — | — | |
| ResNet-152 + GenIntBackbone=ResNet-152, Training Strategy=GenInt Training2021.11 | 60.7 | 59.2 | 60.2 | 62.4 | 70.8 | 59.5 | 69.9 | 64.4 | 63.8 | 58.3 | 48.7 | 61.5 | 70.9 | 55.2 | 60 | 61.7 | — | — | — | — | — | — | — | — | |
| ResNet-50 + CBAMBackbone=ResNet-50, Model Architecture=CBAM (Self-Attention)2020.06 | 60 | — | — | — | 69 | 56 | 61 | — | — | — | — | — | — | — | — | — | — | — | 62 | — | — | — | — | — | |
| Ours (ROID)Backbone=ViT B/16, Source Model Weights=ROID [44]2026.07 | 59.5 | 62 | 61.3 | 60.6 | 61.8 | 55 | 62.2 | 54.4 | 46.6 | 44.8 | 28.5 | 74.2 | 50.5 | 51.5 | 42.2 | 54.3 | — | — | — | — | — | — | — | — | |
| EATAType=BP-based, Setting=continual, Backbone=ViT-Base2026.03 | 59.1 | 61.2 | 64.8 | 65.4 | 60.3 | 64.7 | 63.2 | 68.3 | 68.9 | 72.5 | 79.9 | 60.8 | 67.5 | 73.3 | 72.8 | — | 66.8 | — | — | — | — | — | — | — | |
| ResNet-50 + Speckle NoiseBackbone=ResNet-50, Data Augmentation=Speckle Noise2020.06 | 57 | — | — | — | 68 | 60 | 64 | — | — | — | — | — | — | — | — | — | — | — | 62 | — | — | — | — | — | |
| ResNet-50 + Style TransferBackbone=ResNet-50, Data Augmentation=Style Transfer2020.06 | 57 | — | — | — | 68 | 55 | 64 | — | — | — | — | — | — | — | — | — | — | — | 61 | — | — | — | — | — | |
| FOABackbone=ViT-Large, Number of adaptation samples=5122025.10 | 56.8 | 63.1 | 62.2 | 64 | 46 | 61.8 | 56.8 | 66.6 | 63.9 | 64.7 | 80.1 | 46 | 59 | 75 | 73.7 | — | 62.6 | — | — | — | — | — | — | — | |
| ROIDBackbone=ViT-Base, Evaluation Protocol=Continual Batch Setting2025.10 | 56.6 | 60.68 | 63.04 | 63 | 55.84 | 62.72 | 58.86 | 66.78 | 66.7 | 70.76 | 79.24 | 12.14 | 57.02 | 69.46 | 71.04 | — | 60.93 | — | — | — | — | — | — | — | |
| ZOTTABackbone=quantized ViT-Base, Precision=8-bit floating-point2026.03 | 56.3 | 60.1 | 61.8 | 62 | 56.4 | 61.8 | 57.6 | 69.2 | 66.2 | 70.8 | 80 | 58.8 | 65.8 | 73.3 | 72.5 | — | 64.8 | — | — | — | — | — | — | — | |
| NEOBackbone=ViT-Large, Number of adaptation samples=5122025.10 | 56.2 | 64.5 | 62.9 | 65 | 46.4 | 62.4 | 57 | 68.1 | 64.4 | 67.2 | 80.6 | 42.9 | 59.3 | 75.9 | 74.1 | — | 63.1 | — | — | — | — | — | — | — | |
| ResNet-50 + ImageNet-21K PretrainingBackbone=ResNet-50, Pre-training=ImageNet-21K2020.06 | 56 | — | — | — | 69 | 53 | 59 | — | — | — | — | — | — | — | — | — | — | — | 59 | — | — | — | — | — | |
| DeYOType=BP-based, Setting=continual, Backbone=ViT-Base2026.03 | 56 | 59.8 | 63.5 | 64.6 | 57.7 | 62.1 | 56.4 | 66.5 | 66.2 | 71.7 | 79.3 | 65.6 | 63.8 | 72.1 | 70.9 | — | 65.1 | — | — | — | — | — | — | — | |
| ViT-BBackbone=ViT-B, Pre-training=ImageNet-21K2021.11 | 55.7 | 47 | 48.3 | 46 | 65.6 | 46.7 | 54 | 40 | 40.6 | 32.2 | 42.3 | 42 | 57.1 | 63.1 | 63.6 | 49.62 | — | — | — | — | — | — | — | — | |
| ZOTTAType=BP-free, Setting=continual, Backbone=ViT-Base2026.03 | 55.7 | 62.2 | 64.7 | 64.4 | 57.2 | 61.5 | 57.9 | 65.8 | 67.8 | 69.3 | 79.7 | 64.6 | 62.3 | 71.5 | 71.1 | — | 65 | — | — | — | — | — | — | — | |
| MVT + FTVLM=MMICL, Fine-tuned=true, Severity level=avg2025.12 | 55 | 56.5 | 56.1 | 55.4 | 49.3 | 60.4 | 53.8 | 62.8 | 60.6 | 68.1 | 73.9 | 63.8 | 56.5 | 66.9 | 62.9 | — | — | — | — | — | — | — | — | — | |
| DUSABackbone=ConvNeXt-L, Normalization=Layer normalization, Test-time adaptation protocol=Continual2025.01 | 54.8 | 64.1 | 67.7 | 68.3 | 56.2 | 64.6 | 65.6 | 69.8 | 69.9 | 74.5 | 79 | 70.3 | 68.5 | 71.9 | 70.7 | — | 67.7 | — | — | — | — | — | — | — | |
| DUSABackbone=ConvNeXt-L, Normalization=Layer Normalization, Protocol=Continual Test-Time Adaptation2025.01 | 54.8 | 64.1 | 67.7 | 68.3 | 56.2 | 64.6 | 65.6 | 69.8 | 69.9 | 74.5 | 79 | 70.3 | 68.5 | 71.9 | 70.7 | — | 67.7 | — | — | — | — | — | — | — | |
| DUSABackbone=ConvNeXt-L, Normalization=Layer Normalization, Protocol=Fully Test-Time Adaptation2025.01 | 54.7 | 64.2 | 65.5 | 65.6 | 53.6 | 63.8 | 61.9 | 70.1 | 66.6 | 72.7 | 79.7 | 68.9 | 66.1 | 70.7 | 69.3 | — | 66.2 | — | — | — | — | — | — | — | |
| ViT-B + Dr. ViTBackbone=ViT-B2021.11 | 54.6 | 36.9 | 38.6 | 36 | 57.4 | 53.4 | 63.2 | 45.4 | 38.7 | 34.1 | 40.9 | 39.6 | 56.6 | 45 | 53 | 46.22 | — | — | — | — | — | — | — | — | |
| Machine Vision Therapy (MVT)VLM=MMICL, Severity level=avg2025.12 | 54.5 | 56.4 | 55.7 | 53.9 | 47.3 | 59 | 51.6 | 60.9 | 57.1 | 64.6 | 72.3 | 59.1 | 55 | 65.6 | 63.6 | — | — | — | — | — | — | — | — | — | |
| SARType=BP-based, Setting=continual, Backbone=ViT-Base2026.03 | 54.2 | 59.2 | 61.2 | 61.7 | 55.3 | 58.7 | 55.7 | 61 | 61.9 | 64.6 | 76.8 | 58.4 | 58.4 | 68.5 | 68.8 | — | 61.6 | — | — | — | — | — | — | — | |
| Diffusion-TTABackbone=ConvNeXt-L, Normalization=Layer normalization, Test-time adaptation protocol=Continual2025.01 | 54.1 | 58.1 | 63.2 | 63.2 | 56.6 | 61.8 | 62.5 | 65.2 | 65.5 | 68.1 | 75.3 | 58.9 | 37.3 | 54.8 | 60.9 | — | 60.4 | — | — | — | — | — | — | — | |
| Diffusion-TTABackbone=ConvNeXt-L, Normalization=Layer Normalization, Protocol=Continual Test-Time Adaptation2025.01 | 54.1 | 58.1 | 63.2 | 63.2 | 56.6 | 61.8 | 62.5 | 65.2 | 65.5 | 68.1 | 75.3 | 58.9 | 37.3 | 54.8 | 60.9 | — | 60.4 | — | — | — | — | — | — | — | |
| TENTBackbone=ViT-Large, Number of adaptation samples=5122025.10 | 53.9 | 63.9 | 61.9 | 64.6 | 45.7 | 61.2 | 55.8 | 66.2 | 62.7 | 62.2 | 80.6 | 40.4 | 56.1 | 76.2 | 72.9 | — | 61.6 | — | — | — | — | — | — | — | |
| SurgeonBackbone=ViT-Large, Number of adaptation samples=5122025.10 | 53.5 | 65.6 | 64.6 | 64.1 | 45.5 | 60.5 | 54.9 | 66.1 | 62.2 | 62.1 | 79.7 | 41.1 | 54.7 | 74.4 | 71.1 | — | 61.3 | — | — | — | — | — | — | — | |
| DUSABackbone=ViT-B/16, Normalization=Layer Normalization, Protocol=Fully Test-Time Adaptation2025.01 | 53.3 | 56.6 | 57.9 | 57 | 56.7 | 62.4 | 61.6 | 65.9 | 65.7 | 70.1 | 75.3 | 60.2 | 67.9 | 69.7 | 65.8 | — | 63.1 | — | — | — | — | — | — | — | |
| FOAType=BP-free, Setting=continual, Backbone=ViT-Base2026.03 | 53.2 | 60.8 | 63 | 63.9 | 50.1 | 59.7 | 56.1 | 65.9 | 70 | 70.9 | 80.8 | 65.3 | 53.3 | 72.1 | 72.9 | — | 63.9 | — | — | — | — | — | — | — | |
| SARBackbone=ViT-Large, Number of adaptation samples=5122025.10 | 53.2 | 63.4 | 61.1 | 64.4 | 44.9 | 60.7 | 55.2 | 65.9 | 62.4 | 62.4 | 80.3 | 42.3 | 55.4 | 74.8 | 73 | — | 61.3 | — | — | — | — | — | — | — | |
| T3ABackbone=ViT-Large, Number of adaptation samples=5122025.10 | 53.1 | 63.4 | 61.1 | 64.3 | 44.8 | 60.7 | 55 | 65.8 | 62.3 | 61.9 | 80.5 | 39.2 | 55.7 | 75.6 | 72.7 | — | 61.1 | — | — | — | — | — | — | — | |
| CoTTABackbone=ViT-Large, Number of adaptation samples=5122025.10 | 52.8 | 63.4 | 61.6 | 64.4 | 45.2 | 61 | 55.5 | 66.2 | 62.9 | 62.2 | 80.1 | 39 | 55.5 | 74.7 | 72.2 | — | 61.1 | — | — | — | — | — | — | — | |
| FOABackbone=quantized ViT-Base, Precision=8-bit floating-point2026.03 | 52.7 | 58.3 | 59 | 60.6 | 36 | 54 | 46.5 | 63 | 62.2 | 71 | 77.6 | 55.4 | 45.7 | 68.9 | 68.2 | — | 58.6 | — | — | — | — | — | — | — | |
| No AdaptBackbone=ViT-Large, Number of adaptation samples=5122025.10 | 52.7 | 63.1 | 61.6 | 63.7 | 44.8 | 60.5 | 55 | 66.3 | 62.3 | 62.6 | 80.2 | 39.7 | 56 | 74.9 | 72.7 | — | 61.1 | — | — | — | — | — | — | — | |
| LAMEBackbone=ViT-Large, Number of adaptation samples=5122025.10 | 52.6 | 63 | 60.7 | 64 | 44.3 | 60.3 | 54.6 | 65.1 | 61.7 | 61.4 | 80.1 | 39.2 | 54.7 | 75.4 | 72.4 | — | 60.6 | — | — | — | — | — | — | — | |
| ResNet-50 + AugMixBackbone=ResNet-50, Data Augmentation=AugMix2020.06 | 52 | — | — | — | 65 | 46 | 51 | — | — | — | — | — | — | — | — | — | — | — | 54 | — | — | — | — | — | |
| CLIPBackbone=ViT-L, Severity level=avg2025.12 | 52 | 54.7 | 53.8 | 51.9 | 44.9 | 57.8 | 49.8 | 60 | 56.2 | 63.2 | 72 | 58.4 | 53.5 | 64.5 | 61.4 | — | — | — | — | — | — | — | — | — | |
| FOABackbone=ViT-Base, Evaluation Protocol=Continual Batch Setting2025.10 | 51.98 | 57.14 | 61.62 | 61.96 | 42.78 | 57.48 | 54.08 | 65.98 | 68.44 | 65.6 | 79.48 | 62.78 | 52.2 | 70.32 | 72.2 | — | 61.6 | — | — | — | — | — | — | — | |
| ELaTTABackbone=ViT-Base, Evaluation Protocol=Continual single-instance2025.10 | 51.82 | 61.78 | 62.4 | 62.98 | 39.5 | 57.74 | 47.44 | 68.92 | 68.52 | 68.72 | 80.8 | 31.74 | 56.2 | 70.3 | 67.94 | — | 59.79 | — | — | — | — | — | — | — | |
| TENTType=BP-based, Setting=continual, Backbone=ViT-Base2026.03 | 51.8 | 57.6 | 59.8 | 60.9 | 49.6 | 59.8 | 53.4 | 64 | 62.7 | 68 | 78.6 | 66.6 | 54.5 | 70 | 69.8 | — | 61.8 | — | — | — | — | — | — | — | |
| BECoTTABackbone=ViT-Base, Evaluation Protocol=Continual Batch Setting2025.10 | 51.64 | 61.82 | 62.52 | 62.27 | 38.1 | 58.16 | 48.65 | 67.79 | 68.82 | 36.77 | 82.2 | 30.69 | 49.38 | 71.99 | 72.11 | — | 57.53 | — | — | — | — | — | — | — | |
| ELaTTABackbone=ViT-Base, Evaluation Protocol=Continual Batch Setting2025.10 | 51.08 | 61.38 | 61.98 | 62.64 | 39.16 | 57.7 | 47.2 | 68.76 | 68.6 | 71.5 | 79.96 | 30.78 | 55.78 | 69.28 | 67.08 | — | 59.53 | — | — | — | — | — | — | — | |
| BalanceBackbone=ViT-Base, Evaluation Protocol=Continual Batch Setting2025.10 | 50.34 | 58.52 | 63.92 | 64.46 | 40.66 | 56.58 | 44.94 | 67.62 | 68.42 | 66.48 | 77.74 | 39.54 | 51.66 | 68.46 | 69.7 | — | 59.27 | — | — | — | — | — | — | — | |
| ResNet-50 + DeepAugmentBackbone=ResNet-50, Data Augmentation=DeepAugment2020.06 | 48 | — | — | — | 60 | 51 | 61 | — | — | — | — | — | — | — | — | — | — | — | 55 | — | — | — | — | — | |
| ZOABackbone=ViT-Base, Evaluation Protocol=Continual Batch Setting2025.10 | 47.98 | 58 | 58.56 | 59.38 | 39.1 | 56.38 | 49.38 | 65.68 | 63.48 | 62.68 | 79.34 | 49.02 | 50.76 | 66.84 | 70.34 | — | 58.46 | — | — | — | — | — | — | — | |
| BECoTTABackbone=ViT-Base, Evaluation Protocol=Continual single-instance2025.10 | 47.6 | 56.46 | 55.86 | 55.9 | 34.8 | 52.34 | 43.34 | 63.3 | 62.26 | 66.34 | 77.88 | 31.6 | 45.64 | 66.08 | 66.36 | — | 55.05 | — | — | — | — | — | — | — | |
| T3ABackbone=ViT-Base, Evaluation Protocol=Continual Batch Setting2025.10 | 47.48 | 58.7 | 61.18 | 61.4 | 48.88 | 58.38 | 50.92 | 62.82 | 62.3 | 64.9 | 78.32 | 60.4 | 54.78 | 67.5 | 69.44 | — | 60.49 | — | — | — | — | — | — | — | |
| ViT-B + Dr. ViTBackbone=ViT-B, Pre-training=ImageNet-21K2021.11 | 47.3 | 30.5 | 30.6 | 29.2 | 54.3 | 44.4 | 49.4 | 34.5 | 31.7 | 25.7 | 34.7 | 33.1 | 52.9 | 39.5 | 43.2 | 38.74 | — | — | — | — | — | — | — | — | |
| NoAdaptSetting=continual, Backbone=ViT-Base2026.03 | 46.9 | 56.8 | 56.8 | 57.5 | 35.6 | 53.1 | 44.8 | 62.2 | 62.5 | 65.7 | 77.7 | 32.6 | 46 | 67 | 67.6 | — | 55.5 | — | — | — | — | — | — | — | |
| No AdaptBackbone=ViT-Base, Evaluation Protocol=Continual single-instance2025.10 | 46.48 | 55.34 | 56.23 | 56.01 | 34.78 | 52.87 | 44.2 | 62.39 | 62.66 | 65.56 | 77.7 | 32.04 | 45.73 | 66.72 | 66.67 | — | 55.03 | — | — | — | — | — | — | — | |
| No AdaptBackbone=ViT-Base, Evaluation Protocol=Continual Batch Setting2025.10 | 46.48 | 55.34 | 56.23 | 56.01 | 34.78 | 52.87 | 44.2 | 62.39 | 62.66 | 65.56 | 77.7 | 32.04 | 45.73 | 66.72 | 66.67 | — | 55.03 | — | — | — | — | — | — | — | |
| LAMEType=BP-free, Setting=continual, Backbone=ViT-Base2026.03 | 46.4 | 56.5 | 56.6 | 57.3 | 34.8 | 52.7 | 44.2 | 58.4 | 61.6 | 63.1 | 77.5 | 24.7 | 44.6 | 66.6 | 67.2 | — | 54.1 | — | — | — | — | — | — | — | |
| NoAdaptBackbone=quantized ViT-Base, Precision=8-bit floating-point2026.03 | 45.9 | 55.3 | 55.1 | 56 | 34.5 | 51.9 | 42.1 | 60.8 | 60.7 | 63.3 | 77.2 | 22.3 | 44.2 | 65.9 | 66.7 | — | 53.5 | — | — | — | — | — | — | — | |
| T3ABackbone=quantized ViT-Base, Precision=8-bit floating-point2026.03 | 45.8 | 55.6 | 55.7 | 55.7 | 34.4 | 51.1 | 41.2 | 59.5 | 61.9 | 66.8 | 76.4 | 45.5 | 43.4 | 65.6 | 67.5 | — | 55.1 | — | — | — | — | — | — | — | |
| SARBackbone=ViT-Base, Evaluation Protocol=Continual single-instance2025.10 | 45.52 | 59.08 | 60.52 | 59.36 | 57.26 | 58.56 | 57.12 | 62.74 | 66.66 | 68.68 | 78.78 | 6.68 | 67.16 | 72.4 | 71.56 | — | 59.47 | — | — | — | — | — | — | — |