Composed Image Retrieval on GeneCIS
17.4Recall@1MagicLens-B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MagicLens-BBackbone=CoCa-B, # Total Params=267M2024.03 | 17.4 | — | — | |
| MagicLens-LBackbone=CoCa-L, # Total Params=613M2024.03 | 16.7 | — | — | |
| MagicLens-LBackbone=CLIP-L, # Total Params=465M2024.03 | 16.3 | — | — | |
| CIReVLBackbone=CLIP-B, # Total Params=12.3B*2024.03 | 15.9 | — | — | |
| CIReVLBackbone=CLIP-L, # Total Params=12.5B*2024.03 | 15.9 | — | — | |
| MagicLens-BBackbone=CLIP-B, # Total Params=166M2024.03 | 15 | — | — | |
| CompoDiffBackbone=CLIP-L, # Total Params=568M2024.03 | 14.9 | — | — | |
| LinCIRBackbone=ViT-H2023.12 | 13.76 | 23.87 | 33.16 | |
| LinCIRBackbone=ViT-G2023.12 | 13.66 | 24.64 | 33.54 | |
| SEARLEBackbone=ViT-H2023.12 | 13.34 | 23.72 | 32.72 | |
| SEARLEBackbone=ViT-G2023.12 | 12.87 | 22.61 | 32.46 | |
| SEARLEBackbone=CLIP-L, # Total Params=442M2024.03 | 12.3 | — | — | |
| SEARLEBackbone=ViT-L2023.12 | 12.26 | 22.11 | 31.3 | |
| LinCIRBackbone=CLIP-L, # Total Params=442M2024.03 | 12.2 | — | — | |
| LinCIRBackbone=ViT-L2023.12 | 12.19 | 22.76 | 32.38 | |
| Pic2WordBackbone=ViT-H2023.12 | 11.89 | 22.17 | 31.94 | |
| Pic2WordBackbone=CLIP-L, # Total Params=429M2024.03 | 11.2 | — | — | |
| Pic2WordBackbone=ViT-L2023.12 | 11.16 | 21.47 | 30.38 | |
| Pic2WordBackbone=ViT-G2023.12 | 10.67 | 20.7 | 29.5 |