Table 5 Head-wise ablation analysis of the proposed method. Removing each head degrades detection accuracy even though all data remain in use (annotations of the removed head were redirected to the others head).
From: A multi-head YOLOv12 with self-supervised pretraining for urinary sediment particle detection
Configuration | Precision (%) | Recall (%) | mAP@50 (%) | mAP@50:95 (%) |
|---|---|---|---|---|
Without casts head | \(72.58{\pm }2.67\) | \(60.68{\pm }2.21\) | \(62.73{\pm }1.23\) | \(45.42{\pm }0.83\) |
Without cells head | \(73.69{\pm }3.09\) | \(60.28{\pm }2.74\) | \(61.98{\pm }1.04\) | \(46.56{\pm }0.98\) |
Without crystals head | \(73.58{\pm }2.39\) | \(61.71{\pm }2.39\) | \(60.38{\pm }1.13\) | \(44.70{\pm }0.87\) |
Without microorganisms head | \(72.69{\pm }2.42\) | \(59.35{\pm }3.29\) | \(59.47{\pm }1.12\) | \(44.52{\pm }0.78\) |
Full (all six heads) | \({76.59}{\pm }{3.29}\) | \({62.45}{\pm }{2.94}\) | \({64.15}{\pm }{1.56}\) | \({47.20}{\pm }{0.72}\) |
