Text-to-Text Retrieval on Crisscrossed Captions (CxC)
61.8R@1ALIGN-L2
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| ALIGN-L2Fine-tuned on=MSCOCO2021.09 | 61.8 | 84.9 | 91.1 | — | |
| MURAL-LARGEFine-tuned on=Multi30K and MSCOCO2021.09 | 57.8 | 83.1 | 90 | 9.4 | |
| ALIGN-BASEFine-tuned on=Multi30K and MSCOCO2021.09 | 50 | 77.3 | 85.9 | 11.5 | |
| MURAL-BASEFine-tuned on=Multi30K and MSCOCO2021.09 | 49.7 | 77.5 | 86 | 11 | |
| ALIGN2021.02 | 45.4 | 66.8 | 75.2 | — | |
| DE T2T+I2T2021.02 | 42.4 | 64.9 | 74 | — | |
| DE-T2T+I2TFine-tuned on=MSCOCO2021.09 | 41.7 | 72.3 | 83 | — | |
| VSRN2021.02 | 41 | 64.8 | 74.5 | — | |
| VSE++2021.02 | 38.7 | 62.3 | 72.2 | — | |
| DE I2T2021.02 | 26 | 47.1 | 57.5 | — | |
| ALIGN2022.09 | — | — | — | 62.5 | |
| DET2T+I2T2022.09 | — | — | — | 60.4 | |
| ERNIE-ViL 2.0training=fine-tuned on COCO2022.09 | — | — | — | 64.7 |