Zero-shot Image Classification on ImageNet V2 (test)
71.5AccuracyREACT (Locked-Text Gated-Image)
Evaluation results
| Method | Links | |
|---|---|---|
| REACT (Locked-Text Gated-Image)Backbone=ViT-L/14, Pretrain Data=WIT-400M, Evaluation Protocol=Zero-shot2023.01 | 71.5 | |
| CLIPBackbone=ViT-L/14, Pretrain Data=WIT-400M, Evaluation Protocol=Zero-shot2023.01 | 69.6 | |
| REACT (Locked-Text Gated-Image)Backbone=ViT-L/14, Pretrain Data=LAION-2B, Evaluation Protocol=Zero-shot2023.01 | 68.9 | |
| OpenCLIPBackbone=ViT-L/14, Pretrain Data=LAION-2B, Evaluation Protocol=Zero-shot2023.01 | 67.9 | |
| REACT (Locked-Text Gated-Image)Backbone=ViT-B/16, Pretrain Data=WIT-400M, Evaluation Protocol=Zero-shot2023.01 | 66.8 | |
| REACT (Locked-Text)Backbone=ViT-B/16, Pretrain Data=WIT-400M, Evaluation Protocol=Zero-shot2023.01 | 64.4 | |
| REACT (Locked-Text Gated-Image)Backbone=ViT-B/16, Pretrain Data=LAION-400M, Evaluation Protocol=Zero-shot2023.01 | 63 | |
| REACT (Locked-Text)Backbone=ViT-B/16, Pretrain Data=LAION-400M, Evaluation Protocol=Zero-shot2023.01 | 62.4 | |
| CLIPBackbone=ViT-B/16, Pretrain Data=WIT-400M, Evaluation Protocol=Zero-shot2023.01 | 61.8 | |
| REACT (Locked-Text Gated-Image)Backbone=ViT-B/32, Pretrain Data=LAION-2B, Evaluation Protocol=Zero-shot2023.01 | 61.6 | |
| REACT (Locked-Text Gated-Image)Backbone=ViT-B/32, Pretrain Data=WIT-400M, Evaluation Protocol=Zero-shot2023.01 | 61 | |
| REACT (Locked-Text)Backbone=ViT-B/32, Pretrain Data=LAION-2B, Evaluation Protocol=Zero-shot2023.01 | 59.5 | |
| OpenCLIPBackbone=ViT-B/16, Pretrain Data=LAION-400M, Evaluation Protocol=Zero-shot2023.01 | 59.4 | |
| REACT (Locked-Text Gated-Image)Backbone=ViT-B/32, Pretrain Data=LAION-400M, Evaluation Protocol=Zero-shot2023.01 | 58.7 | |
| REACT (Locked-Text)Backbone=ViT-B/32, Pretrain Data=WIT-400M, Evaluation Protocol=Zero-shot2023.01 | 58.6 | |
| OpenCLIPBackbone=ViT-B/32, Pretrain Data=LAION-2B, Evaluation Protocol=Zero-shot2023.01 | 58.2 | |
| REACT (Locked-Text)Backbone=ViT-B/32, Pretrain Data=LAION-400M, Evaluation Protocol=Zero-shot2023.01 | 57.3 | |
| CLIPBackbone=ViT-B/32, Pretrain Data=WIT-400M, Evaluation Protocol=Zero-shot2023.01 | 55.9 | |
| OpenCLIPBackbone=ViT-B/32, Pretrain Data=LAION-400M, Evaluation Protocol=Zero-shot2023.01 | 55.2 |