Recognizing the Damaged Surface Parts of Cars in the Real Scene Using a Deep Learning Framework

Table 4

Quantitative comparison between the proposed model and eight other models based on recall, precision, and IoU for detecting different damaged parts of a car.

Architecture	Precision (%)	Recall (%)	IoU
PANet [42]	87	86	0.85
Combined feature(YOLOv3) [43]	67	68	0.64
CNN [47]	58	61	0.57
FFNN [48]	88	89	0.86
Mask RCNN [45]	84	81	0.78
VGG models [1]	85	84	0.81
HTC [46]	89	90	0.87
Texture descriptor [8]	90	91	0.88
Proposed method	93	92	0.90