Vision-Language Diagnostic Evaluation on COCO Flickr30K pool
46.04Caption R@1PCA prefix
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| PCA prefixNumber of Parameters=02026.05 | 46.04 | 39.24 | -2.51 | 59.09 | 0 | |
| MRL-styleNumber of Parameters=262,1492026.05 | 41.44 | 29.77 | 0.94 | 40.24 | 0 | |
| SMEC-style compressionNumber of Parameters=262,1492026.05 | 41.11 | 39.08 | -3.24 | 57.84 | 0 | |
| Matryoshka-style adaptorNumber of Parameters=294,9182026.05 | 38.97 | 29.78 | 1.08 | 40.96 | 0 | |
| Frozen fullNumber of Parameters=02026.05 | 36.91 | — | — | 74.05 | 0 | |
| GraSP-VLNumber of Parameters=262,1492026.05 | 34.91 | 53.01 | 33.17 | 89.76 | 0 | |
| Learned permutationNumber of Parameters=262,1492026.05 | 31.91 | 29.56 | 2.01 | 47.65 | 0 | |
| Learned signed permutationNumber of Parameters=262,6612026.05 | 29.25 | 27.27 | -1.97 | 43.6 | 0 | |
| Generic MLP adapterNumber of Parameters=1,052,1652026.05 | 24.72 | 50.88 | 37.5 | 88.24 | 0.36 | |
| Random rotationNumber of Parameters=02026.05 | 23.31 | 37.34 | — | 66.57 | 0 | |
| Direct prefixNumber of Parameters=02026.05 | 14.46 | 26.48 | 7.84 | 46.17 | 0 |