Table 1.

AlphaTracker demonstrated superior performance in tracking both two mice in a home cage and four mice in an operant chamber.

	Two mice home cage			Four mice operant chamber
	MOTA	MOTP	mAP	MOTA	MOTP	mAP
AlphaTracker	82.2	86.2	87.2	84.0	87.2	85.6
DeepLabCut	40.6	77.6	14.1	71.4	86.5	68.0
SLEAP	73.6	76.7	26.8	77.9	86.5	83.9

AlphaTracker, DeepLabCut, and SLEAP were evaluated on two datasets: two mice in a home cage and four mice in an operant chamber. Each model was trained on 600 annotated frames and evaluated on 200 frames with human-labeled ground truth. The evaluation results show that AlphaTracker outperforms DeepLabCut and SLEAP in both datasets, achieving higher keypoint detection accuracy (mAP) and tracking consistency (MOTA and MOTP).