Visual Question Answering on RSVQA-HR (test)
79.2HR Presence ScoreGRASP
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GRASPParams (M)=4.252026.01 | 79.2 | 82.3 | 74.1 | 65.3 | 69.5 | |
| DoRAParams (M)=50.22026.01 | 79 | 82.1 | 73.6 | 64.8 | 68.5 | |
| LoRAParams (M)=50.12026.01 | 78.6 | 81.8 | 73.3 | 64.2 | 68.6 | |
| VPTParams (M)=6.72026.01 | 78.5 | 80.1 | 71.9 | 61.5 | 67.3 | |
| AdapterParams (M)=16.72026.01 | 78.4 | 80.9 | 72.8 | 63.1 | 68.8 | |
| APromptParams (M)=16.12026.01 | 71.9 | 77.2 | 66.3 | 58.6 | 57.4 | |
| Prompt TuningParams (M)=0.412026.01 | 70.2 | 76.7 | 64.6 | 57.4 | 53.9 | |
| LLaVA-1.52026.02 | 69.83 | 67.29 | 68.56 | — | — | |
| Qwen2-VL-HFmitigation strategy=RSHalluShield, fine-tuning data=object-level questions2026.02 | 67.63 | 75.52 | 71.58 | — | — | |
| Qwen2-VL2026.02 | 67.08 | 70.29 | 68.88 | — | — | |
| Qwen-VL-Chatmode=Chat2026.02 | 66.44 | 60.41 | 63.43 | — | — | |
| LLaVA-Next (Base)2026.01 | 66.4 | 75.6 | 61.8 | 56 | 49.1 | |
| VHM2026.02 | 64 | 83.5 | 73.75 | — | — | |
| Gemini-Vision2026.02 | 63.6 | 64.6 | 64.1 | — | — | |
| EarthGPT2026.02 | 62.77 | 79.53 | 71.15 | — | — | |
| GeoChat2026.02 | 58.45 | 83.19 | 70.82 | — | — | |
| MiniGPTv22026.02 | 40.79 | 50.91 | 45.85 | — | — |