Table 2.

CNN architecture details for Task 2

	Name	Type	Activations	Learnable Parameters	Total Learnables
1	227×227×3 input images	Input image	227×227×3	-	0
2	96 11×11×3 convolutions with stride [4 4] and padding [0 0 0 0]	Convolution	55×55×96	Weights: 11×11×3×96 Bias: 1×1×96	34,944
3	ReLU-1	ReLU	55×55×96	-	0
4	Cross channel normalization with 5 channels per element	Cross-channel normalization	55×55×96	-	0
5	3×3 maxpooling with stride [2 2] and padding [0 0 0 0]	Maxpooling	27×27×96	-	0
6	2 groups of 128 5×5×48 convolutions with stride [1 1] and padding [2 2 2 2]	Grouped convolution	27×27×256	Weights: 5×5×48×128×2 Bias: 1×1×128×2	307,456
7	ReLU-2	ReLU	27×27×256	-	0
8	3×3 maxpooling with stride [2 2] and padding [0 0 0 0]	Maxpooling	13×13×256	-	0
9	2 groups of 128 5×5×48 convolutions with stride [1 1] and padding [2 2 2 2]	Grouped convolution	27×27×256	Weights: 5×5×48×128×128×2 Bias: 1×1×128×2	307,456
10	ReLU-3	ReLU	13×13×256	-	0
11	3×3 maxpooling with stride [2 2] and padding [0 0 0 0]	Maxpooling	6×6×256	-	0
12	3 fully connected layer	Fully connected	1×1×3	Weights: 3×9216 Bias: 3×1	27,651
13	Softmax	Softmax	1×1×2	-	0
14	Crossentropyex with classes Normal, ‘Pneumonia, and COVID-19	classification output	-	-	0