Region-level Visual Grounding on DVGBench 1.0 (test)
57.14SecurityDroneVG-R1 7B
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| DroneVG-R1 7BRationale=true, backbone_size=7B, Algorithm=GRPO2026.01 | 57.14 | 52.08 | 45.05 | 43.4 | 70.66 | 43.95 | 52.05 | |
| Qwen2.5-VL 7BRationale=false, SFT=true, backbone_size=7B2026.01 | 54.76 | 48.61 | 45.05 | 41.51 | 66.53 | 45.22 | 50.28 | |
| DroneVG-R1 3BRationale=true, backbone_size=3B, Algorithm=GRPO2026.01 | 50 | 42.36 | 41.76 | 30.19 | 57.85 | 40.13 | 43.71 | |
| Qwen2.5-VL 3BRationale=false, SFT=true, backbone_size=3B2026.01 | 45.24 | 20.14 | 41.76 | 47.17 | 45.45 | 40.13 | 39.98 | |
| Qwen2.5-VL 3BRationale=true, backbone_size=3B2026.01 | 45.24 | 25.69 | 29.67 | 22.64 | 51.24 | 35.03 | 34.92 | |
| InternVL3.5 4BRationale=true, backbone_size=4B2026.01 | 42.86 | 14.58 | 14.29 | 16.98 | 17.36 | 18.47 | 20.76 | |
| Qwen2.5-VL 7BRationale=true, backbone_size=7B2026.01 | 35.71 | 41.32 | 34.07 | 47.17 | 63.64 | 40.76 | 43.78 | |
| LQVGRationale=false, SFT=true2026.01 | 33.33 | 23.61 | 16.48 | 15.09 | 43.39 | 19.75 | 25.28 | |
| InternVL3.5 8BRationale=true, backbone_size=8B2026.01 | 33.33 | 28.47 | 27.47 | 26.42 | 31.4 | 25.48 | 28.76 | |
| DeepSeek-VL2-TinyRationale=true, backbone_size=Tiny2026.01 | 21.43 | 8.33 | 16.48 | 15.09 | 31.82 | 24.2 | 19.56 | |
| GeoGroundRationale=false, SFT=false2026.01 | 19.05 | 18.06 | 29.67 | 20.75 | 33.88 | 11.46 | 22.15 | |
| GeoChatRationale=false, SFT=false2026.01 | 11.91 | 2.43 | 8.79 | 9.43 | 4.55 | 3.82 | 6.82 | |
| MGVLFRationale=false, SFT=false2026.01 | 7.14 | 0 | 13.19 | 5.66 | 4.96 | 3.18 | 5.69 | |
| MGVLFRationale=false, SFT=true2026.01 | 4.76 | 0.35 | 5.49 | 0 | 20.66 | 1.91 | 5.53 | |
| LQVGRationale=false, SFT=false2026.01 | 2.38 | 13.54 | 9.89 | 1.89 | 6.2 | 8.28 | 7.03 | |
| LHRS-BotRationale=false, SFT=false2026.01 | 0 | 1.04 | 2.2 | 0 | 2.07 | 5.1 | 1.73 |