Wide Baseline Matching on ReasonMatch-Bench Outdoor (test)
90.9L1 ScoreQwen3-VL-8B + DCRL
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3-VL-8B + DCRLBackbone=Qwen3-VL-8B, Method=DCRL2026.06 | 90.9 | 80.2 | 73.6 | |
| GPT-5-mini2026.06 | 75.8 | 75 | 51.4 | |
| Gemini-2.5-Pro2026.06 | 63.4 | 64.5 | 46.6 | |
| GPT-5-Chat2026.06 | 61.8 | 75.1 | 44.9 | |
| Qwen3-VL-235BParameters=235B2026.06 | 58.8 | 71.1 | 41.7 | |
| Claude-4.5-Sonnet2026.06 | 50.3 | 56.9 | 37.4 | |
| Claude-4.1-Opus2026.06 | 45.1 | 45.3 | 31.6 | |
| Claude-4-Sonnet2026.06 | 45 | 54.5 | 29.8 | |
| GPT-4o-2411062026.06 | 33.9 | 54 | 23.3 | |
| Qwen3-VL-8B-InstructParameters=8B, Mode=Instruct2026.06 | 30.4 | 39.5 | 18.8 |