Visual Question Answering on OK-VQA (Robustness Accuracy)
59.6Accuracy (Clean)UniTok-MLLM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| UniTok-MLLMTokenizer=clean2026.02 | 59.6 | 2.1 | 0.8 | |
| UniTok-MLLMTokenizer=AT 4/2552026.02 | 53 | 38.8 | 29.9 | |
| UniTok-MLLMTokenizer=AT 8/2552026.02 | 49.2 | 40.3 | 32.8 | |
| UniTok-MLLMTokenizer=AT 12/2552026.02 | 46.7 | 39.4 | 33.9 | |
| UniTok-MLLMTokenizer=AT 16/2552026.02 | 45.1 | 38.8 | 32.2 |