Test-Time Adaptation on ImageNet-C level 5 (test)
66.6AccuracyZOTTA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ZOTTABackbone=ViT-Base, Rely on BP?=No, Learn from an objective?=Yes, Learnable params=Norm layer, Model compatibility=ViTs & CNNs, etc., Batch size=64, Precision=Full-precision2026.03 | 66.6 | 825 | |
| SARBackbone=ViT-Base, Rely on BP?=Yes, Learn from an objective?=Yes, Learnable params=Norm layer, Model compatibility=ViTs, CNNs, etc., Batch size=64, Precision=Full-precision2026.03 | 62.7 | 5,166 | |
| FOABackbone=ViT-Base, Rely on BP?=No, Learn from an objective?=Yes, Learnable params=Prompt, Model compatibility=ViTs, Batch size=64, Precision=Full-precision2026.03 | 62.4 | 830 | |
| T3ABackbone=ViT-Base, Rely on BP?=No, Learn from an objective?=No, Learnable params=N/A, Model compatibility=ViTs, CNNs, etc., Batch size=64, Precision=Full-precision2026.03 | 56.9 | 957 | |
| SARBackbone=ResNet50-GN, Rely on BP?=Yes, Learn from an objective?=Yes, Learnable params=Norm layer, Model compatibility=ViTs, CNNs, etc., Batch size=64, Precision=Full-precision2026.03 | 46.8 | 5,429 | |
| ZOTTABackbone=ResNet50-GN, Rely on BP?=No, Learn from an objective?=Yes, Learnable params=Norm layer, Model compatibility=ViTs & CNNs, etc., Batch size=64, Precision=Full-precision2026.03 | 44.9 | 782 | |
| T3ABackbone=ResNet50-GN, Rely on BP?=No, Learn from an objective?=No, Learnable params=N/A, Model compatibility=ViTs, CNNs, etc., Batch size=64, Precision=Full-precision2026.03 | 30.9 | 969 |