Visual Question Answering on MVQA
30.46LLM-J Score (Qwen2.5-7B)KG-ViP
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| KG-ViPReasoning Strategy=Retrieval-Augmented Reasoning2026.01 | 30.46 | 35.44 | 18.44 | 23.25 | |
| LLM-RAReasoning Strategy=Retrieval-Augmented Reasoning2026.01 | 22.53 | 24.1 | 11.43 | 17.22 | |
| FilterRAGReasoning Strategy=Retrieval-Augmented Reasoning2026.01 | 20.11 | 20.84 | 12.41 | 15.49 | |
| NaiveRAGReasoning Strategy=Retrieval-Augmented Reasoning2026.01 | 19.47 | 20.78 | 9.11 | 13.17 | |
| CCoTReasoning Strategy=Scene Graph–Augmented Reasoning2026.01 | 16.41 | 18.64 | 8.69 | 10.23 | |
| Fine-TuningReasoning Strategy=Vanilla MLLM Reasoning2026.01 | 16.23 | 16.85 | 8.33 | 11.22 | |
| CoTReasoning Strategy=Vanilla MLLM Reasoning2026.01 | 16.09 | 17.78 | 8.2 | 9.69 | |
| Zero-shotReasoning Strategy=Vanilla MLLM Reasoning2026.01 | 14.41 | 15.84 | 7.7 | 9.18 |