Image-to-Text Retrieval on Crisscrossed Captions (CxC)
78.1R@1ALIGN
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| ALIGN2021.02 | 78.1 | 94.3 | 97.4 | — | |
| VACSRBackbone=CLIP ViT-B/162026.05 | 73.3 | 92 | 96.2 | — | |
| PCME++Backbone=CLIP ViT-B/162026.05 | 69.9 | 91.3 | 95.7 | — | |
| VACSRBackbone=CLIP ViT-B/322026.05 | 67.8 | 89.6 | 94.9 | — | |
| PCMEBackbone=CLIP ViT-B/162026.05 | 66.8 | 90.5 | 95.4 | — | |
| PCME++Backbone=CLIP ViT-B/322026.05 | 63.5 | 88.4 | 94 | — | |
| DAABackbone=CLIP ViT-B/322026.05 | 61.5 | 86.8 | 93.3 | — | |
| PCMEBackbone=CLIP ViT-B/322026.05 | 61.5 | 87.5 | 93.5 | — | |
| P2RMBackbone=CLIP ViT-B/162026.05 | 58.5 | 86 | 92.7 | — | |
| P2RMBackbone=CLIP ViT-B/322026.05 | 58.1 | 85.5 | 92.2 | — | |
| DE T2T+I2T2021.02 | 55.9 | 84.2 | 91.8 | — | |
| DE I2T2021.02 | 53.9 | 82.7 | 91.2 | — | |
| VSRN2021.02 | 52.4 | 81.9 | 90 | — | |
| MURAL-LARGEFine-tuned on=Multi30K and MSCOCO2021.09 | 46.5 | 67.5 | 76.1 | 47.8 | |
| ALIGN-L2Fine-tuned on=MSCOCO2021.09 | 45.4 | 66.8 | 75.2 | — | |
| MURAL-BASEFine-tuned on=Multi30K and MSCOCO2021.09 | 43.9 | 64.9 | 73.9 | 44.9 | |
| ALIGN-BASEFine-tuned on=Multi30K and MSCOCO2021.09 | 43.5 | 64.7 | 73.5 | 45.4 | |
| VSE++2021.02 | 43.1 | 74.3 | 84.2 | — | |
| DE-T2T+I2TFine-tuned on=MSCOCO2021.09 | 42.4 | 64.9 | 74 | — | |
| DAABackbone=CLIP ViT-B/162026.05 | 26.4 | 53.9 | 67.1 | — | |
| ALIGN2022.09 | — | — | — | 89.9 | |
| DET2T+I2T2022.09 | — | — | — | 77.3 | |
| ERNIE-ViL 2.0training=fine-tuned on COCO2022.09 | — | — | — | 90 |