Text-to-Image Retrieval on UCM (test)
22.71R@1MRS-Adapter
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| MRS-AdapterBackbone=CLIP(ViT-B-32), Params=0.16M, Method Category=CLIP-based methods2023.08 | 22.71 | 55.81 | 80.33 | 55.71 | |
| Single LanguageBackbone=CLIP(ViT-B-32), Params=151M, Method Category=CLIP-based methods2023.08 | 19.04 | 53.33 | 77.61 | 54.12 | |
| LW-MCR-uBackbone=SqueezeNet, Params=1.65M, Method Category=Traditional methods2023.08 | 18.1 | 47.14 | 63.81 | 45.35 | |
| Full fine-tuningBackbone=CLIP(ViT-B-32), Params=151M, Method Category=CLIP-based methods2023.08 | 17.14 | 55.24 | 79.52 | 52.43 | |
| AdaptFormerBackbone=CLIP(ViT-B-32), Params=0.17M, Method Category=CLIP-based methods2023.08 | 16.92 | 54.39 | 77.46 | 53.69 | |
| AMFMN-fusionBackbone=ResNet18, biGRU, Params=35.94M, Method Category=Traditional methods2023.08 | 16.67 | 45.71 | 68.57 | 46.08 | |
| ConvpassBackbone=CLIP(ViT-B-32), Params=0.14M, Method Category=CLIP-based methods2023.08 | 16.46 | 54.79 | 78.78 | 52.91 | |
| LW-MCR-dBackbone=SqueezeNet, Params=1.65M, Method Category=Traditional methods2023.08 | 15.24 | 51.9 | 62.86 | 44.68 | |
| CABIRBackbone=ResNet152, BERT+biGRU, Params=15.17M, Method Category=Traditional methods2023.08 | 15.17 | 45.71 | 72.85 | 48.3 | |
| CAMP-bceBackbone=Faster R-CNN, biGRU, Params=27.03M, Method Category=Traditional methods2023.08 | 14.76 | 46.19 | 67.62 | 43.92 | |
| AMFMN-simBackbone=ResNet18, biGRU, Params=35.94M, Method Category=Traditional methods2023.08 | 14.76 | 49.52 | 68.1 | 45.68 | |
| UniAdapterBackbone=CLIP(ViT-B-32), Params=0.55M, Method Category=CLIP-based methods2023.08 | 14.46 | 55.19 | 83.95 | 54.59 | |
| SCAN t-iBackbone=Faster R-CNN, biGRU, Params=13.68M, Method Category=Traditional methods2023.08 | 14.29 | 45.71 | 67.62 | 44.67 | |
| Cross-Modal AdapterBackbone=CLIP(ViT-B-32), Params=0.16M, Method Category=CLIP-based methods2023.08 | 13.77 | 50.57 | 81.41 | 53.62 | |
| Linear probeBackbone=CLIP(ViT-B-32), Params=1.05M, Method Category=CLIP-based methods2023.08 | 13.33 | 52.71 | 77.62 | 51.3 | |
| AMFMN-softBackbone=ResNet18, biGRU, Params=35.94M, Method Category=Traditional methods2023.08 | 12.86 | 51.9 | 66.67 | 45.97 | |
| SCAN i-tBackbone=Faster R-CNN, biGRU, Params=13.68M, Method Category=Traditional methods2023.08 | 12.85 | 47.14 | 69.52 | 43.43 | |
| VSE++Backbone=ResNet18, GRU, Params=15.78M, Method Category=Traditional methods2023.08 | 12.38 | 44.76 | 65.71 | 36.93 | |
| LW-MCR-bBackbone=SqueezeNet, Params=1.65M, Method Category=Traditional methods2023.08 | 12.38 | 43.81 | 59.52 | 41.1 | |
| CAMP-tripletBackbone=Faster R-CNN, biGRU, Params=27.03M, Method Category=Traditional methods2023.08 | 10.95 | 44.29 | 65.71 | 42.22 | |
| Zero-shot CLIPBackbone=CLIP(ViT-B-32), Params=0.00M, Method Category=CLIP-based methods2023.08 | 10.95 | 34.76 | 55.71 | 33.08 | |
| MTFNBackbone=Faster R-CNN, biGRU, Params=77.90M, Method Category=Traditional methods2023.08 | 10.47 | 47.62 | 64.29 | 44.65 | |
| AdapterBackbone=CLIP(ViT-B-32), Params=0.17M, Method Category=CLIP-based methods2023.08 | 10.17 | 49.73 | 80.08 | 52.74 | |
| CLIP-AdapterBackbone=CLIP(ViT-B-32), Params=0.52M, Method Category=CLIP-based methods2023.08 | 9.83 | 42.52 | 66.79 | 44.82 | |
| VPTBackbone=CLIP(ViT-B-32), Params=0.46M, Method Category=CLIP-based methods2023.08 | 8.86 | 47.81 | 78.62 | 49.77 | |
| VL-PromptBackbone=CLIP(ViT-B-32), Params=0.47M, Method Category=CLIP-based methods2023.08 | 7.97 | 45.48 | 75.96 | 47.5 | |
| CoOpBackbone=CLIP(ViT-B-32), Params=0.10M, Method Category=CLIP-based methods2023.08 | 7.19 | 42.76 | 74.19 | 44.95 |