Referring Expression Retrieval on RefCOCO (val)
61.2Acc@1MaskInversion
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MaskInversionBackbone=ViT-H/14, zero-shot=true2024.07 | 61.2 | 61.8 | 47.5 | |
| MaskInversionBackbone=ViT-B/16, zero-shot=true2024.07 | 56.1 | 56.8 | 44.5 | |
| MaskInversionBackbone=ViT-L/14, zero-shot=true2024.07 | 56.1 | 56.7 | 42 | |
| AlphaCLIPBackbone=ViT-B/16 + ViT-L/14, zero-shot=false, Deviating evaluation settings=true2024.07 | 55.7 | — | — | |
| MaskInversionBackbone=ViT-B/32, zero-shot=true2024.07 | 54.1 | 54.7 | 42.3 | |
| FGVPBackbone=RN50x16 + ViT-B/32 + ViT-L/14@336, zero-shot=true, Deviating evaluation settings=true2024.07 | 52.9 | — | — | |
| Masked CropBackbone=ViT-B/16, zero-shot=true, Reproduced=true2024.07 | 52.3 | 52.9 | 41.2 | |
| RedCircleBackbone=RN50x16 + ViT-L/14@336, zero-shot=true, Deviating evaluation settings=true2024.07 | 49.8 | — | — | |
| ReCLIPBackbone=RN50x16 + ViT-B/32, zero-shot=true, Deviating evaluation settings=true2024.07 | 45.8 | — | — | |
| AlphaCLIPBackbone=ViT-B/16, zero-shot=false, Reproduced=true2024.07 | 43.4 | 44 | 38.1 | |
| GradCAMBackbone=RN50x16 + ViT-B/32, zero-shot=true, Deviating evaluation settings=true2024.07 | 42.9 | — | — | |
| FGVPBackbone=ViT-B/16, zero-shot=true, Reproduced=true2024.07 | 42.6 | 43.2 | 33.3 | |
| RedCircleBackbone=ViT-B/16, zero-shot=true, Reproduced=true2024.07 | 42.5 | 43.2 | 32.7 | |
| CPTBackbone=RN50x16 + ViT-B/32, zero-shot=true, Deviating evaluation settings=true2024.07 | 32.2 | — | — | |
| CLIPBackbone=ViT-B/16, zero-shot=true, Reproduced=true2024.07 | 18.3 | 18.9 | 15.3 | |
| CropBackbone=ViT-B/16, zero-shot=true, Reproduced=true2024.07 | 17.9 | 18.5 | 15.5 | |
| RISBackbone=ViT-B/32, zero-shot=true2024.07 | — | — | 42.6 |