Wide Baseline Matching on ReasonMatch-Bench Object (test)
45.6L1 MetricQwen3-VL-8B + DCRL
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3-VL-8B + DCRLBackbone=Qwen3-VL-8B, Method=DCRL2026.06 | 45.6 | 63.1 | 33.7 | |
| GPT-5-mini2026.06 | 45.3 | 60.4 | 27.8 | |
| GPT-5-Chat2026.06 | 44.1 | 57.2 | 28.6 | |
| Qwen3-VL-235BParameters=235B2026.06 | 35.1 | 49.4 | 22.1 | |
| Claude-4.5-Sonnet2026.06 | 33.9 | 43.4 | 19.4 | |
| GPT-4o-2411062026.06 | 31.9 | 42.2 | 19.9 | |
| Claude-4-Sonnet2026.06 | 30.8 | 44.1 | 18.3 | |
| Gemini-2.5-Pro2026.06 | 29.5 | 35.9 | 19 | |
| Claude-4.1-Opus2026.06 | 26.3 | 37.2 | 16.1 | |
| Qwen3-VL-8B-InstructParameters=8B, Mode=Instruct2026.06 | 21.1 | 34.5 | 11.5 |