Remote Sensing Visual Question Answering on RSVQA-LR unified
31.1CountGRASP
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GRASPTrainable Parameters (M)=4.25, Category=Prompt-based tuning methods2026.01 | 31.1 | 0.826 | 0.816 | 0.815 | 0.692 | |
| VPTTrainable Parameters (M)=6.70, Category=Prompt-based tuning methods2026.01 | 30.3 | 0.809 | 0.793 | 0.793 | 0.675 | |
| DoRATrainable Parameters (M)=50.2, Category=Additive parameter tuning methods2026.01 | 30 | 0.829 | 0.812 | 0.811 | 0.688 | |
| LoRATrainable Parameters (M)=50.1, Category=Additive parameter tuning methods2026.01 | 29.5 | 0.83 | 0.808 | 0.813 | 0.687 | |
| AdapterTrainable Parameters (M)=16.7, Category=Additive parameter tuning methods2026.01 | 27.6 | 0.814 | 0.79 | 0.805 | 0.671 | |
| APromptTrainable Parameters (M)=16.1, Category=Prompt-based tuning methods2026.01 | 23.4 | 0.781 | 0.725 | 0.73 | 0.618 | |
| Prompt TuningTrainable Parameters (M)=0.41, Category=Prompt-based tuning methods2026.01 | 22.3 | 0.778 | 0.714 | 0.729 | 0.611 | |
| Qwen2.5-VL (Base)2026.01 | 20.9 | 0.66 | 0.623 | 0.714 | 0.552 |