Visual Question Answering on Flickr30k RoQA (test)
80.43BERTScore PQwen2-VL-RoVQA
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Qwen2-VL-RoVQAModel Size=7B, Adaptation Strategy=RoVQA Adapter2025.12 | 80.43 | 80.15 | 80.15 | 48.1 | 48.63 | 44.25 | 67.86 | |
| LLaMA-3.2-RoVQAModel Size=11B, Adaptation Strategy=RoVQA Adapter2025.12 | 75.97 | 75.95 | 75.8 | 37.01 | 37.74 | 32.82 | 77.11 | |
| Qwen2-VLModel Size=7B, Adaptation Strategy=None (Base Model)2025.12 | 71.59 | 77.11 | 74.1 | 27.11 | 45.44 | 30.22 | 29.12 | |
| LLaMA-3.2Model Size=11B, Adaptation Strategy=None (Base Model)2025.12 | 61 | 69.08 | 64.64 | 8.48 | 29.15 | 10.51 | 18.51 | |
| LLaVA-v1.6Model Size=7B, Adaptation Strategy=None (Base Model)2025.12 | 58.17 | 63.31 | 60.33 | 9.11 | 8.21 | 6.38 | 16.6 | |
| LLaVA-v1.6-RoVQAModel Size=7B, Adaptation Strategy=RoVQA Adapter2025.12 | 49.5 | 63.01 | 55.13 | 4.21 | 11.13 | 3.98 | 50.39 |