Area Estimation on RS-GSD Benchmark v5.0
17.1Median ErrorClaude Opus 4.6
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Claude Opus 4.6vehicle-length hint=cars ≈ 5 m, operation mode=agent with programmatic image analysis tools2026.03 | 17.1 | 41.5 | 36 | 59 | 73 | 88 | |
| RS-GSDmode=proposed pipeline2026.03 | 19.7 | 29.2 | 30 | 59 | 80 | 97 | |
| GPT-4ovehicle-length hint=cars ≈ 5 m2026.03 | 35.9 | 55.2 | 15 | 36 | 59 | 85 | |
| Qwen2.5-VL-72Bvehicle-length hint=cars ≈ 5 m2026.03 | 38.1 | 100.4 | 10 | 30 | 59 | 83 | |
| Qwen2.5-VL-72Bscale information=none, inference mode=Zero-shot2026.03 | 38.3 | 84.5 | 19 | 35 | 61 | 84 | |
| Qwen-VL-Maxvehicle-length hint=cars ≈ 5 m2026.03 | 46.5 | 99.9 | 13 | 30 | 53 | 80 | |
| Claude Opus 4.6scale information=none, inference mode=Zero-shot2026.03 | 47.2 | 72.9 | 24 | 35 | 55 | 73 | |
| Qwen-VL-Maxscale information=none, inference mode=Zero-shot2026.03 | 50.3 | 97.1 | 13 | 27 | 49 | 75 | |
| Qwen-VL-Plusscale information=none, inference mode=Zero-shot2026.03 | 51.3 | 190.3 | 22 | 33 | 49 | 68 | |
| GPT-4oscale information=none, inference mode=Zero-shot2026.03 | 51.9 | 69.4 | 17 | 34 | 48 | 93 | |
| Qwen-VL-Plusvehicle-length hint=cars ≈ 5 m2026.03 | 58 | 236.2 | 12 | 28 | 44 | 70 |