Image-to-Image Retrieval on Crisscrossed Captions (CxC)
50.3R@1MURAL-LARGE
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| MURAL-LARGEFine-tuned on=Multi30K and MSCOCO2021.09 | 50.3 | 81.8 | 90.1 | 12.4 | |
| ALIGN2021.02 | 49.4 | 81.4 | 89.1 | — | |
| ALIGN-L2Fine-tuned on=MSCOCO2021.09 | 49.4 | 81.4 | 89.1 | — | |
| VSRN2021.02 | 44.2 | 76.7 | 86.2 | — | |
| MURAL-BASEFine-tuned on=Multi30K and MSCOCO2021.09 | 43.9 | 76.7 | 86.5 | 16.1 | |
| ALIGN-BASEFine-tuned on=Multi30K and MSCOCO2021.09 | 42.6 | 76.6 | 86.2 | 16 | |
| DE T2T+I2T2021.02 | 38.5 | 73.6 | 84.9 | — | |
| DE-T2T+I2TFine-tuned on=MSCOCO2021.09 | 38.5 | 73.6 | 84.9 | — | |
| DE I2T2021.02 | 38.3 | 74.1 | 85 | — | |
| VSE++2021.02 | 36.4 | 70.4 | 81.3 | — | |
| ALIGN2022.09 | — | — | — | 73.3 | |
| DET2T+I2T2022.09 | — | — | — | 65.7 | |
| ERNIE-ViL 2.0training=fine-tuned on COCO2022.09 | — | — | — | 72.4 |