Pairwise taxonomic verification accuracy on Birds-to-Words
79.4Visual AccuracyTaxonRL
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| TaxonRL2026.03 | 79.4 | 83.7 | 91.7 | 100 | 100 | 100 | 91.7 | |
| GRPO2026.03 | 72.1 | 83.7 | 89.6 | 100 | 100 | 100 | 89.8 | |
| Neural Naturalist2026.03 | 55 | 45 | 67.5 | 70 | 72.5 | 77.5 | 64.6 | |
| SFTTraining=Supervised Fine-Tuning2026.03 | 52.9 | 85.7 | 43.8 | 75.5 | 92.2 | 95.5 | 72.8 | |
| DinoV2_GiantModel Size=Giant2026.03 | 52.7 | 69.3 | 59.1 | 70.3 | 85.9 | 100 | 76.9 | |
| Human2026.03 | 52.5 | 40.2 | 88.7 | 98.4 | 98 | 100 | 77.3 | |
| Qwen2.5-VL-7BParameters=7B2026.03 | 51.4 | 86.8 | 41.5 | 70.9 | 90.7 | 93.8 | 70.9 |