Count on Count (test)
99.8AccuracyFinetuning + KL
Evaluation Results
| Method | Links | |
|---|---|---|
| Finetuning + KL2025.05 | 99.8 | |
| Tracr Injection2025.05 | 99.2 | |
| Ours (multi-object)consistent-checkpoint=true2026.05 | 72.97 | |
| VisionReasoner-7Bconsistent-checkpoint=true2026.05 | 69.5 | |
| Qwen2.5-VL-7Bre-implemented=true, consistent-checkpoint=true2026.05 | 67.9 | |
| Qwen2-VL-7Bre-implemented=true, consistent-checkpoint=true2026.05 | 48 | |
| No editing2025.05 | 0 |