deepghs/real_head_detection

These models are trained on deepghs/anime_head_detection and open-sourced real photos datasets.

So both anime and photos are supported.

Use this with dghs-realutils

pip install dghs-realutils

from realutils.detect import detect_heads

print(detect_heads('yolo/solo.jpg'))
# [((162, 47, 305, 210), 'head', 0.7701659202575684)]
print(detect_heads('yolo/2girls.jpg'))
# [((683, 48, 1199, 754), 'head', 0.8410779237747192), ((105, 91, 570, 734), 'head', 0.8339194059371948)]
print(detect_heads('yolo/3+cosplay.jpg'))
# [((329, 194, 426, 309), 'head', 0.8123012781143188), ((359, 20, 448, 122), 'head', 0.8047150373458862), ((185, 81, 265, 166), 'head', 0.7797152996063232)]
print(detect_heads('yolo/multiple.jpg'))
# [((867, 259, 1084, 527), 'head', 0.8264595866203308), ((1364, 448, 1583, 724), 'head', 0.8254891633987427), ((480, 201, 781, 565), 'head', 0.8191508054733276), ((1189, 175, 1398, 412), 'head', 0.8097156286239624), ((1028, 671, 1277, 992), 'head', 0.8084591627120972)]

For more information, see documentation of realutils.

Model	Type	FLOPS	Params	F1 Score	Threshold	precision(B)	recall(B)	mAP50(B)	mAP50-95(B)	F1 Plot	Confusion	Labels
head_detect_v0_l_yv11	yolo	87.3G	25.3M	0.81	0.199	0.90226	0.72872	0.81049	0.5109	plot	confusion	head
head_detect_v0_m_yv11	yolo	68.2G	20.1M	0.8	0.206	0.89855	0.72654	0.80704	0.50804	plot	confusion	head
head_detect_v0_s_yv11	yolo	21.5G	9.43M	0.78	0.187	0.88726	0.69234	0.77518	0.47825	plot	confusion	head
head_detect_v0_n_yv11	yolo	6.44G	2.59M	0.74	0.14	0.87359	0.64011	0.73393	0.44118	plot	confusion	head
head_detect_v0_l	yolo	165G	43.6M	0.81	0.234	0.89921	0.74092	0.81715	0.51615	plot	confusion	head
head_detect_v0_m	yolo	79.1G	25.9M	0.8	0.228	0.90006	0.72646	0.80614	0.50586	plot	confusion	head
head_detect_v0_s	yolo	28.6G	11.1M	0.78	0.182	0.89224	0.69382	0.77804	0.48067	plot	confusion	head
head_detect_v0_n	yolo	8.19G	3.01M	0.74	0.172	0.8728	0.64823	0.73865	0.44501	plot	confusion	head