Generalist Multi-task Evaluation on ImageNet-1K and COCO
-11.8Mean DeltaCLIP
Evaluation Results
| Method | Links | |
|---|---|---|
| CLIPVariant=B, # Train=13B, FLOPS % (Per Iter.)=100, FLOPS % (Total)=322024.06 | -11.8 | |
| CLIPVariant=L, # Train=13B, FLOPS % (Per Iter.)=100, FLOPS % (Total)=322024.06 | -11 | |
| OpenCLIPVariant=L, # Train=32B, FLOPS % (Per Iter.)=100, FLOPS % (Total)=802024.06 | -6.3 | |
| LessIsMoreVariant=B, # Train=11B, FLOPS % (Per Iter.)=100, FLOPS % (Total)=282024.06 | -5.9 | |
| OpenCLIPVariant=B, # Train=34B, FLOPS % (Per Iter.)=100, FLOPS % (Total)=852024.06 | -5.8 | |
| EVA-CLIPVariant=B, # Train=8B, FLOPS % (Per Iter.)=100, FLOPS % (Total)=202024.06 | -4.6 | |
| EVA-CLIPVariant=L, # Train=4B, FLOPS % (Per Iter.)=100, FLOPS % (Total)=102024.06 | -3.4 | |
| SigLIPVariant=B, # Train=40B, FLOPS % (Per Iter.)=100, FLOPS % (Total)=1002024.06 | 0 | |
| SigLIPVariant=L, # Train=40B, FLOPS % (Per Iter.)=100, FLOPS % (Total)=1002024.06 | 0 | |
| SILC-SVariant=B, # Train=20B, FLOPS % (Per Iter.)=380, FLOPS % (Total)=1902024.06 | 0.2 | |
| Flexi-JEST++Variant=B, # Train=4B, FLOPS % (Per Iter.)=110, FLOPS % (Total)=112024.06 | 0.9 | |
| JEST++Variant=L, # Train=4B, FLOPS % (Per Iter.)=233, FLOPS % (Total)=232024.06 | 1.8 | |
| JEST++Variant=B, # Train=4B, FLOPS % (Per Iter.)=233, FLOPS % (Total)=232024.06 | 2.8 |