Visual Counting on VIEW2SPACE
91.37AccuracyHuman-max
Evaluation results
| Method | Links | |
|---|---|---|
| Human-max2026.03 | 91.37 | |
| Human-avg2026.03 | 86.24 | |
| Human-min2026.03 | 81.05 | |
| Ours (Grounded CoT)Model Type=Fine-tuned Model, Backbone=Qwen3VL-4B2026.03 | 54.99 | |
| GPT-52026.03 | 38.1 | |
| Qwen3-VL-4B2026.03 | 21.15 | |
| Random (frequency)2026.03 | 13.05 | |
| Random (chance)2026.03 | 3.39 |