Text-rich Image Question Answering (Abstract) on TRINS-VQA (test)
45.3B@1LLaVAR (finetuned)
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| LLaVAR (finetuned)Resolution=336^2, Fine-tuning=true2024.06 | 45.3 | 31.9 | 23.9 | 18.7 | 20.4 | 39.6 | 104.7 | |
| LaRAResolution=336^22024.06 | 45.1 | 34.2 | 27.3 | 22.6 | 21.9 | 46.5 | 186.6 | |
| LLaVAR w/ OCRResolution=336^2, OCR integration=true2024.06 | 40.3 | 28 | 20.9 | 16.3 | 18.7 | 37.3 | 97.3 | |
| Qwen-VLResolution=448^22024.06 | 40.2 | 27.3 | 19.9 | 15.2 | 20.4 | 36.4 | 79.4 | |
| LLaVARResolution=336^22024.06 | 38 | 25.9 | 19.1 | 14.7 | 17.4 | 35.6 | 84.9 | |
| mPLUG-Owl2Resolution=448^22024.06 | 35.9 | 25.5 | 19.3 | 15.2 | 17.7 | 38 | 97.2 | |
| LLaVAResolution=336^22024.06 | 33.1 | 20.9 | 14.4 | 10.5 | 18 | 31.8 | 48.2 | |
| LLaVA 1.5Resolution=336^22024.06 | 29.8 | 20.9 | 15.8 | 12.6 | 20.8 | 34.9 | 40.7 | |
| Mini-GPTv2Resolution=224^22024.06 | 29.4 | 19.4 | 13.8 | 10.4 | 17.2 | 28.7 | 44.3 | |
| Instruct-BLIPResolution=224^22024.06 | 13.5 | 9.2 | 6.7 | 5.2 | 8.9 | 16.7 | 23.5 |