Image Captioning on MSCOCO (val)
139.1CIDErEVLGen
Evaluation Results
| Method | Links | |
|---|---|---|
| EVLGenzero-shot=true, # pre-train image-text=104M, # trainable params=110M, # stage-1 steps=-, # stage-2 steps=250k, Clock time=133 hrs2023.10 | 139.1 | |
| BLIP-2zero-shot=true, # pre-train image-text=104M, # trainable params=110M+, # stage-1 steps=250k, # stage-2 steps=80k, Clock time=234 hrs2023.10 | 137.7 | |
| EVLGenzero-shot=true, # pre-train image-text=104M, # trainable params=110M, # stage-1 steps=-, # stage-2 steps=150k, Clock time=80 hrs2023.10 | 137 | |
| EVLGenzero-shot=true, # pre-train image-text=11M, # trainable params=110M, # stage-1 steps=-, # stage-2 steps=150k, Clock time=80 hrs2023.10 | 135.1 | |
| EVLGenzero-shot=true, # pre-train image-text=104M, # trainable params=55M, # stage-1 steps=-, # stage-2 steps=90k, Clock time=47 hrs2023.10 | 134 |