Human Correlation for Image Captioning Evaluation on EX-8K
60.2Kendall's tau-cBEiTScore-L/14
Evaluation Results
| Method | Links | |
|---|---|---|
| BEiTScore-L/14Size (M)=674.69, GFLOPs=296.732026.05 | 60.2 | |
| BEiTScore-B/32Size (M)=222.28, GFLOPs=83.852026.05 | 59.4 | |
| BEiTScore-L/14 w/o Stage 1Size (M)=674.69, GFLOPs=296.732026.05 | 57.8 | |
| BEiTScore-B/32 w/o Stage 1Size (M)=222.28, GFLOPs=83.852026.05 | 57.6 | |
| PAC-S++/ViT-LSize (M)=430.58, GFLOPs=56.262026.05 | 57.4 | |
| EXPERTSize (M)=13350.84, GFLOPs=9317.472026.05 | 56.7 | |
| PAC-S/ViT-LSize (M)=427.62, GFLOPs=56.262026.05 | 55.7 | |
| CLIP-S/ViT-LSize (M)=427.62, GFLOPs=84.362026.05 | 54.7 | |
| PAC-S++/ViT-BSize (M)=155.42, GFLOPs=4.882026.05 | 54.5 | |
| PAC-S/ViT-BSize (M)=151.28, GFLOPs=4.882026.05 | 54.3 | |
| LongCLIP-S/ViT-LSize (M)=427.75, GFLOPs=98.92026.05 | 53.9 | |
| FLEURSize (M)=13350.84, GFLOPs=9317.472026.05 | 53 | |
| LongCLIP-S/ViT-BSize (M)=149.84, GFLOPs=26.242026.05 | 53 | |
| CLIP-S/ViT-BSize (M)=151.28, GFLOPs=7.282026.05 | 52.3 | |
| VQAScore (CLIP-FlanT5-XXL)Size (M)=11459.59, GFLOPs=3742.12026.05 | 41 | |
| SPECSSize (M)=151.49, GFLOPs=9.22026.05 | 30.4 |