Infrared Image Deblurring Based on Generative Adversarial Networks

Table 1

The layer structure and parameters of the generator.

Layer (type)	Output shape	Parameters
ReflectionPad2d-1	[−1, 3, 262, 262]	0
Conv2d-2	[−1, 64, 256, 256]	1,792
InstanceNorm2d-3	[−1, 64, 256, 256]	0
ReLU-4	[−1, 64, 256, 256]	0
Conv2d-5	[−1, 128, 128, 128]	73,856
InstanceNorm2d-6	[−1, 128, 128, 128]	0
ReLU-7	[−1, 128, 128, 128]	0
Conv2d-8	[−1, 256, 64, 64]	295,168
InstanceNorm2d-9	[−1, 256, 64, 64]	0
ReLU-10	[−1, 256, 64, 64]	0
ReflectionPad2d-11	[−1, 256, 66, 66]	0
Conv2d-12	[−1, 256, 64, 64]	590,080
ReLU-13	[−1, 256, 64, 64]	0
ReflectionPad2d-14	[−1, 256, 66, 66]	0
Conv2d-15	[−1, 256, 64, 64]	590,080
ResidualBlock-16	[−1, 256, 64, 64]	0
ConvTranspose2d-65	[−1, 128, 128, 128]	295,040
InstanceNorm2d-66	[−1, 128, 128, 128]	0
ReLU-67	[−1, 128, 128, 128]	0
ConvTranspose2d-68	[−1, 64, 256, 256]	73,792
InstanceNorm2d-69	[−1, 64, 256, 256]	0
ReLU-70	[−1, 64, 256, 256]	0
ReflectionPad2d-71	[−1, 64, 262, 262]	0
Conv2d-72	[−1, 3, 256, 256]	1,731
Tanh-73	[−1, 3, 256, 256]	0