Text-Rich VQA on TextVQA (Score)
80.2ScoreScaled base VLM 7B + Joint FFT
Evaluation results
| Method | Links | |
|---|---|---|
| Scaled base VLM 7B + Joint FFTTrain Data=3.6M + 1.6M2026.06 | 80.2 | |
| Scaled base VLM 7B + MERIT (Proposed, 2D)Train Data=3.6M + 1.6M2026.06 | 79.8 | |
| Base VLM 7B + Joint FFTTrain Data=0.7M + 1.6M2026.06 | 73.8 | |
| Base VLM 7B + MERIT (Proposed, 2D)Train Data=0.7M + 1.6M2026.06 | 72.4 | |
| Scaled base VLM 7BTrain Data=3.6M2026.06 | 72.4 | |
| LLaVA-1.5-13BTrain Data=0.7M2026.06 | 61.3 | |
| LLaVA-1.5-7BTrain Data=0.7M2026.06 | 58.2 | |
| Base VLM 7BTrain Data=0.7M2026.06 | 57.6 |