Table 6 Evaluation results of YOLO12-Nano with different input image sizes on the OpenUrine test set. Metrics are reported as mean±std (%) across five cross-validation folds. The table demonstrates how increasing input resolution substantially improves the detection of urinary sediment particles, with the best performance achieved at \(960 \times 960\) pixels.
From: A multi-head YOLOv12 with self-supervised pretraining for urinary sediment particle detection
Input size | Precision (%) | Recall (%) | mAP50 (%) | mAP50-95 (%) |
|---|---|---|---|---|
40 | \(79.60{\pm }10.1\) | \(7.75{\pm }1.74\) | \(9.73{\pm }0.15\) | \(5.40{\pm }0.18\) |
80 | \(43.86{\pm }3.82\) | \(20.41{\pm }1.80\) | \(15.07{\pm }0.18\) | \(9.51{\pm }0.54\) |
160 | \(54.90{\pm }6.82\) | \(29.39{\pm }0.84\) | \(28.94{\pm }0.56\) | \(19.83{\pm }0.15\) |
320 | \(53.70{\pm }0.76\) | \(38.09{\pm }1.44\) | \(39.51{\pm }0.30\) | \(27.52{\pm }0.47\) |
640 | \(60.55{\pm }1.69\) | \(51.57{\pm }0.63\) | \(51.88{\pm }0.07\) | \(36.92{\pm }0.41\) |
960 | \({{\bf 63.40}}{\pm } {{\bf 0.30}}\) | \({{\bf 53.00}}{\pm } {{\bf 2.10}}\) | \({{\bf 55.98}}{\pm } {{\bf 0.31}}\) | \({{\bf 38.91}}{\pm } {{\bf 0.68}}\) |
1280 | \(56.29{\pm }3.97\) | \(53.49{\pm }3.08\) | \(54.00{\pm }0.30\) | \(36.44{\pm }0.45\) |
