Image Classification on ImageNet 21K (test)
25.3Top-1 AccuracyPOMP
Evaluation results
| Method | Links | |
|---|---|---|
| POMPBackbone=ViT-B/16, Training dataset=ImageNet-21K2023.04 | 25.3 | |
| VPTBackbone=ViT-B/16, Training dataset=ImageNet-21K2023.04 | 24.8 | |
| MaPLeBackbone=ViT-B/16, Training dataset=ImageNet-1K2023.04 | 24.2 | |
| Prompt EnsembleBackbone=ViT-B/16, Protocol=zero-shot inference2023.04 | 23.5 | |
| POMPBackbone=ViT-B/32, Training dataset=ImageNet-21K2023.04 | 22.2 | |
| ZeroshotCLIPBackbone=ViT-B/16, Protocol=zero-shot inference2023.04 | 21.8 | |
| VPTBackbone=ViT-B/32, Training dataset=ImageNet-21K2023.04 | 21.8 | |
| MaPLeBackbone=ViT-B/32, Training dataset=ImageNet-1K2023.04 | 21.6 | |
| Prompt EnsembleBackbone=ViT-B/32, Protocol=zero-shot inference2023.04 | 20.9 | |
| Linear ProbingBackbone=ViT-B/16, Training dataset=ImageNet-21K2023.04 | 20.9 | |
| CoOpBackbone=ViT-B/16, Training dataset=ImageNet-1K2023.04 | 20.8 | |
| POMPBackbone=ResNet50, Training dataset=ImageNet-21K2023.04 | 20.2 | |
| ZeroshotCLIPBackbone=ViT-B/32, Protocol=zero-shot inference2023.04 | 19.8 | |
| Prompt EnsembleBackbone=ResNet50, Protocol=zero-shot inference2023.04 | 18.8 | |
| Linear ProbingBackbone=ViT-B/32, Training dataset=ImageNet-21K2023.04 | 18.2 | |
| CoOpBackbone=ViT-B/32, Training dataset=ImageNet-1K2023.04 | 18.1 | |
| ZeroshotCLIPBackbone=ResNet50, Protocol=zero-shot inference2023.04 | 17.5 | |
| CoOpBackbone=ResNet50, Training dataset=ImageNet-1K2023.04 | 16.6 | |
| Linear ProbingBackbone=ResNet50, Training dataset=ImageNet-21K2023.04 | 6.5 |