Visual Question Answering on ViTextVQA (test)
53.95F1 ScoreViTextBLIP-2
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ViTextBLIP-2Version=base, Method Type=ViT-LM, Model Size=≈2.5B, % Param Tuning=≈5%2024.04 | 53.95 | 25.48 | |
| PreSTUVersion=large, Method Type=ViT-LM, Model Size=≈860M, % Param Tuning=≈100%2024.04 | 44.93 | 22.64 | |
| PreSTUVersion=base, Method Type=ViT-LM, Model Size=≈315M, % Param Tuning=≈100%2024.04 | 43.81 | 20.85 | |
| SaLVersion=base, Method Type=CNN-LM, Model Size=≈270M, % Param Tuning=≈100%2024.04 | 43.49 | 20.11 | |
| SaLVersion=large, Method Type=CNN-LM, Model Size=≈780M, % Param Tuning=≈100%2024.04 | 43.39 | 21.48 | |
| LaTrVersion=large, Method Type=ViT-LM, Model Size=≈900M, % Param Tuning=≈100%2024.04 | 42.34 | 20.94 | |
| LaTrVersion=base, Method Type=ViT-LM, Model Size=≈330M, % Param Tuning=≈100%2024.04 | 40.41 | 17.85 | |
| BLIP-2Version=large, Method Type=ViT-LM, Model Size=≈1.4B, % Param Tuning=≈8%2024.04 | 38.47 | 15.58 | |
| BLIP-2Version=base, Method Type=ViT-LM, Model Size=≈450M, % Param Tuning=≈25%2024.04 | 36.87 | 13.59 | |
| M4CVersion=base, Method Type=CNN-LM, Model Size=≈140M, % Param Tuning=≈100%2024.04 | 25.46 | 10.26 |