Pedestrian Image Captioning on CUHK-PEDES (test)
32.9B@4BLIP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BLIPBackbone=ViT-B, Params=446.0M, PD=129M, TM=true2023.12 | 32.9 | 103.8 | |
| BLIP-2Backbone=ViT-G+, Params=1.1B, PD=129M, TM=true2023.12 | 32.8 | 105.8 | |
| mPLUG-BBackbone=ViT-B, Params=350.0M, PD=14M, TM=true2023.12 | 32.2 | 101.1 | |
| HulkBackbone=ViT-L, Params=336.0M, PD=12M, TM=false, Type=Generalist2023.12 | 31.6 | 94.5 | |
| OFA-LBackbone=ResNet-101, Params=472.0M, PD=18M, TM=false2023.12 | 31.5 | 97.7 | |
| HulkBackbone=ViT-B, Params=113.7M, PD=12M, TM=false, Type=Generalist2023.12 | 31.1 | 91.4 | |
| Hulk-FTBackbone=ViT-L, Params=336.0M, PD=12M, TM=false, Type=Generalist, Mode=Fine-tuned2023.12 | 30.5 | 82.8 | |
| Hulk-FTBackbone=ViT-B, Params=113.7M, PD=12M, TM=false, Type=Generalist, Mode=Fine-tuned2023.12 | 28.3 | 72.8 |