Image Classification on ImageNet and OOD Variants (ImV2, Im-A, Im-R, Im-S) 1.0 (val)
0.8419ImNet AccFTP
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| FTPBackbone=CLIP ViT-Base, Pre-training=CLIP, Protocol=Fine-tuning2023.10 | 0.8419 | 0.7464 | 0.265 | 0.4723 | 0.5023 | 0.4965 | 0.5656 | |
| vanilla FTBackbone=CLIP ViT-Base, Pre-training=CLIP, Protocol=Fine-tuning2023.10 | 0.8366 | 0.7382 | 0.214 | 0.4306 | 0.4552 | 0.4698 | 0.5429 | |
| L2-SPBackbone=CLIP ViT-Base, Pre-training=CLIP, Protocol=Fine-tuning2023.10 | 0.8344 | 0.732 | 0.2055 | 0.4389 | 0.466 | 0.4606 | 0.5354 | |
| LP-FTBackbone=CLIP ViT-Base, Pre-training=CLIP, Protocol=Fine-tuning2023.10 | 0.8299 | 0.7296 | 0.2108 | 0.4465 | 0.4756 | 0.4656 | 0.5385 | |
| WISE-FTPBackbone=CLIP ViT-Base, Pre-training=CLIP, Protocol=Linear Interpolation2023.10 | 0.8261 | 0.7409 | 0.3456 | 0.6118 | 0.5506 | 0.5622 | 0.615 | |
| WISE-FTBackbone=CLIP ViT-Base, Pre-training=CLIP, Protocol=Linear Interpolation2023.10 | 0.8094 | 0.7247 | 0.3318 | 0.6333 | 0.542 | 0.5558 | 0.6082 | |
| Linear Prob.Backbone=CLIP ViT-Base, Pre-training=CLIP, Protocol=Linear Probing2023.10 | 0.7825 | 0.6768 | 0.2654 | 0.5257 | 0.4826 | 0.4876 | 0.5466 | |
| zero-shotBackbone=CLIP ViT-Base, Pre-training=CLIP, Protocol=zero-shot2023.10 | 0.6768 | 0.6141 | 0.306 | 0.5677 | 0.4553 | 0.4858 | 0.524 |