Image-Text Retrieval on WIT Under-resourced v1 (test)
0.729Recall@1 (tg)MURAL-LARGE
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| MURAL-LARGEEvaluation Protocol=Fine-tuned2021.09 | 0.729 | 0.535 | 0.514 | 0.698 | 0.623 | 0.823 | 0.767 | 84.2 | |
| MURAL-BASEEvaluation Protocol=Fine-tuned2021.09 | 0.642 | 0.441 | 0.419 | 0.593 | 0.551 | 0.764 | 0.676 | 79 | |
| ALIGN-BASEEvaluation Protocol=Fine-tuned2021.09 | 0.53 | 0.363 | 0.358 | 0.503 | 0.45 | 0.724 | 0.625 | 78.1 | |
| ALIGN-BASE-ENEvaluation Protocol=Fine-tuned2021.09 | 0.47 | 0.296 | 0.327 | 0.377 | 0.442 | 0.484 | 0.535 | — | |
| MURAL-LARGEEvaluation Protocol=Zero-shot2021.09 | 0.457 | 0.337 | 0.308 | 0.334 | 0.456 | 0.456 | 0.524 | 37.7 | |
| MURAL-BASEEvaluation Protocol=Zero-shot2021.09 | 0.353 | 0.241 | 0.208 | 0.214 | 0.33 | 0.357 | 0.391 | 26.1 | |
| ALIGN-BASE-ENEvaluation Protocol=Zero-shot2021.09 | 0.3 | 0.234 | 0.261 | 0.273 | 0.336 | 0.349 | 0.416 | — | |
| ALIGN-BASEEvaluation Protocol=Zero-shot2021.09 | 0.212 | 0.156 | 0.129 | 0.089 | 0.239 | 0.31 | 0.331 | 24 |