Edge Detection/Classification on ReasoningFlow manually annotated set (test)
64.6F1 ScoreGemini-3.1-Pro
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Gemini-3.1-ProDecoding strategy=greedy decoding, Segmentation type=ground-truth segmentation2026.06 | 64.6 | 0.677 | |
| Gemini-3-FlashDecoding strategy=greedy decoding, Segmentation type=ground-truth segmentation2026.06 | 58.3 | 0.607 | |
| GPT-5.1Decoding strategy=greedy decoding, Segmentation type=ground-truth segmentation2026.06 | 57.4 | 0.604 | |
| Gemini-2.5-FlashDecoding strategy=greedy decoding, Segmentation type=ground-truth segmentation2026.06 | 53.5 | 0.576 | |
| GPT-5-miniDecoding strategy=greedy decoding, Segmentation type=ground-truth segmentation2026.06 | 42.5 | 0.444 |