Text-rich image question-answering on TRINS-VQA Human
58.8AccuracyLaRA
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| LaRAResolution=336^22024.06 | 58.8 | 31 | 23.6 | 18.6 | 15.1 | 26.5 | 38 | 135.6 | |
| LLaVAR (finetuned)Resolution=336^22024.06 | 50.1 | 26.1 | 18.7 | 14.1 | 11 | 23.1 | 33.4 | 112.6 | |
| LLaVARResolution=336^22024.06 | 44.1 | 27.5 | 19.5 | 14.4 | 10.9 | 21.6 | 33 | 94.7 | |
| LLaVAResolution=336^22024.06 | 22.6 | 12.3 | 7.5 | 4.9 | 3.4 | 15.3 | 20.2 | 26.2 |