Glass Refraction Distortion Object Detection via Abstract Features

Table 5

A number of parameters of different algorithms and average detection results in multiple environments (black font indicates the best data, underlined font indicates second best, Ours-1 freezes distorted image reconstruction module).

Algorithms	FLOPs	Parameters	Average Acc	Average IoU
Deep face	2.7 G	114.44 M	79.31%	0.77
VGG face	1.6 G	131.94 M	80.64%	0.76
TBE-CNN	1.75 G	12.29 M	84.14%	0.81
DA-GAN	2.5 G	27.38 M	88.00%	0.85
PEN-3D	2.55 G	4.46 M	84.06%	0.81
LMZMPM	0.72 G	59.48 M	91.16%	0.85
Ours	1.44 G	5.13 M	92.57%	0.88
Ours-1	—	—	85.24%	0.80