PFLO: a high-throughput pose estimation model for field maize based on YOLO architecture

Table 8 The results of the module ablation experiments

Group	SE-layer	RepPose detection head	Dynamic upsample	Multi-SEAM layer	Precision (%)		Recall (%)		mAP50 (%)		mAP50 - 95 (%)
Group	SE-layer	RepPose detection head	Dynamic upsample	Multi-SEAM layer	Box	Pose	Box	Pose	Box	Pose	Box	Pose
A(Baseline)					85.9	72.8	85.1	68.9	90.7	69.9	63.2	39.8
B	√				85.9	73.7	85.7	69.1	90.9**	70.5	63.6*	40.5*
C		√			86.3	73.5	85.4	69.1	91.0	70.5*	63.7*	40.1*
D			√		86.5	73.7	85.2	68.9	90.9	70.2*	63.9	40.4*
E				√	86.8	74.5	86.0	70.1	91.3*	71.2**	64.6*	40.5*
F	√	√	√	√	86.8	75.2	86.2	70.2	91.6*	72.2*	66.0*	42.7*

ISSN: 1746-4811