Image Retrieval on MS COCO 5K
60.2R@1SoftMask++
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SoftMask++#Img=4M, fine-tuned=true2023.04 | 60.2 | 83.7 | 90.5 | — | |
| ALIGN#Img=1.2B, fine-tuned=true2023.04 | 59.9 | 83.3 | 89.8 | — | |
| TCL#Img=4M, fine-tuned=true2023.04 | 59 | 83.2 | 89.9 | — | |
| VinVL#Img=6M, fine-tuned=true2023.04 | 58.8 | 83.5 | 90.3 | — | |
| CODIS#Img=4M, fine-tuned=true2023.04 | 58.7 | 82.8 | 89.7 | — | |
| ALBEF#Img=4M, fine-tuned=true2023.04 | 56.8 | 81.5 | 89.2 | — | |
| OSCAR#Img=4M, fine-tuned=true2023.04 | 54 | 80.8 | 88.5 | — | |
| UNITER#Img=4M, fine-tuned=true2023.04 | 52.9 | 79.9 | 88 | — | |
| SOHO#Img=200K, fine-tuned=true2023.04 | 50.6 | 78 | 86.7 | — | |
| ViLT#Img=4M, fine-tuned=true2023.04 | 42.7 | 72.9 | 83.1 | — | |
| VSRN*split=5K2020.03 | 40.5 | 70.6 | 81.1 | 415.7 | |
| Full-IMRAMvariant=Full model, split=5K2020.03 | 39.7 | 69.1 | 79.8 | 416.5 | |
| Image-IMRAMvariant=RAM block grounded on images, split=5K2020.03 | 38.9 | 68.5 | 79.2 | 412.7 | |
| SCAN*split=5K2020.03 | 38.6 | 69.3 | 80.4 | 410.9 | |
| Text-IMRAMvariant=RAM block grounded on texts, split=5K2020.03 | 38.6 | 68.1 | 79.1 | 409.7 | |
| SCOsplit=5K2020.03 | 33.1 | 62.9 | 75.5 | 369.6 | |
| PVSEsplit=5K2020.03 | 32.4 | 63 | 75 | 374.4 | |
| DPCsplit=5K2020.03 | 25.3 | 53.4 | 66.4 | 337.9 | |
| CYCLIPEvaluation Protocol=Fine-tuned2022.05 | 11.4 | 25.8 | 33.4 | — | |
| CLIPEvaluation Protocol=Fine-tuned2022.05 | 10.6 | 23.9 | 31.3 | — | |
| CYCLIPEvaluation Protocol=Zero-shot2022.05 | 8.6 | 20 | 27 | — | |
| CLIPEvaluation Protocol=Zero-shot2022.05 | 8.4 | 19.5 | 26.6 | — |