TCM: Temporal Consistency Model for Head Detection in Complex Videos

Table 1

Performance evaluation on HollywoodHeads dataset.

Detectors	Baseline CNN	mAP	TCM (mAP)	Improvement (%age)
Faster R-CNN [15]	ZF [42]	0.76	0.83	9.20%
	VGG16 [41]	0.79	0.85	7.59%
	VGGM [41]	0.78	0.81	3.84%
R-CNN [12]	ZF [42]	0.71	0.79	11.26%
	VGG16 [41]	0.74	0.83	12.16%
	VGGM [41]	0.72	0.81	12.67%
YOLO [14]	13-layered architecture	0.39	0.46	18.42%
SSD [13]	VGG16 [41]	0.43	0.54	25.58%
			Average	12.59%