Glass Refraction Distortion Object Detection via Abstract Features

Table 1

Detection results of different algorithms in GRI dataset for glass reflection images (black font indicates the best data, underlined font indicates second best).

Algorithms	Normal		Bright light		Low light		Occlude		Blurred		Average
	Acc(%)	IoU	Acc(%)	IoU	Acc(%)	IoU	Acc(%)	IoU	Acc(%)	IoU	Acc(%)	IoU
Deep face	85.82	0.82	81.27	0.75	82.95	0.77	86.02	0.83	83.42	0.79	83.90	0.79
VGG face	83.26	0.81	80.73	0.77	82.81	0.79	83.59	0.82	78.65	0.75	81.81	0.79
TBE-CNN	91.68	0.84	88.36	0.73	91.08	0.82	90.17	0.83	89.73	0.81	90.20	0.80
DA-GAN	92.71	0.90	93.59	0.86	91.96	0.87	92.24	0.91	90.79	0.85	92.26	0.86
PEN-3D	90.85	0.85	87.61	0.78	88.55	0.78	89.74	0.85	86.51	0.74	88.65	0.79
LMZMPM	95.85	0.87	92.71	0.85	92.15	0.82	93.35	0.87	91.17	0.80	93.05	0.83
Ours	95.57	0.92	93.38	0.91	94.92	0.91	93.31	0.90	92.24	0.88	93.88	0.88