Glass Refraction Distortion Object Detection via Abstract Features

Table 2

Glass refraction distortion image detection results of different algorithms in the GRI dataset (black font indicates the best data, underlined font indicates second best).

Algorithms	Partially distorted		Facial distortion		Regular distortion		Irregular distortion		Color filter		Occlude		Average
	Acc(%)	IoU	Acc(%)	IoU	Acc(%)	IoU	Acc(%)	IoU	Acc(%)	IoU	Acc(%)	IoU	Acc(%)	IoU
Deep face	73.26	0.81	71.53	0.75	72.17	0.73	70.68	0.72	75.42	0.75	68.59	0.70	71.94	0.74
VGG face	77.65	0.72	76.15	0.71	74.52	0.75	72.43	0.74	78.35	0.76	71.61	0.72	75.12	0.73
TBE-CNN	81.28	0.78	79.63	0.76	78.35	0.77	75.61	0.73	80.53	0.79	77.16	0.76	78.76	0.77
DA-GAN	86.92	0.83	83.42	0.81	84.28	0.82	81.82	0.79	85.19	0.81	85.07	0.79	84.45	0.81
PEN-3D	84.47	0.71	82.61	0.84	81.26	0.71	80.56	0.70	83.07	0.73	82.65	0.75	82.44	0.74
LMZMPM	87.29	0.75	85.96	0.78	84.17	0.76	83.29	0.75	93.45	0.79	93.04	0.76	87.87	0.77
Ours	93.17	0.91	91.08	0.83	90.15	0.83	84.31	0.81	92.62	0.82	92.83	0.80	90.69	0.83