Table 2.

Layers’ organization of the proposed DDCAE. The output shape refers to the tuple $\left(M_l,N_l,F_l\right)$ for the $l$th layer.

Layer type	Kernel size	Stride	Output Shape	Param. #
Conv2D	3 × 3	1	(200, 300, 256)	7168
Batch Normalization	–	–	(200, 300, 256)	1024
Conv2D	3 × 3	1	(200, 300, 128)	295040
MaxPooling	2 × 2	2	(100, 150, 128)	0
Conv2D	3 × 3	1	(100, 150, 64)	73792
Batch Normalization	–	–	(100, 150, 64)	256
MaxPooling	2 × 2	2	(50, 75, 64)	0
Total params: 377,280
Trainable params: 376,640
Non-trainable params: 640
Layer type	Kernel size	Stride	Output Shape	Param. #
Conv2D Transpose	3 × 3	2	(100, 150, 128)	73856
Batch Normalization	–	–	(100, 150, 128)	512
Conv2D Transpose	3 × 3	2	(200, 300, 256)	295168
Batch Normalization	–	–	(200, 300, 256)	1024
Conv2D Transpose	3 × 3	1	(200, 300, 3)	6915
Total params: 377,475
Trainable params: 376,707
Non-trainable params: 768