How to calculate output shape & parameters in a CNN?

n_{o u t} = ⌊ \frac{n_{i n} + 2 p - k}{s} ⌋ + 1

Parameters = (k_{w} \times k_{h} \times C_{i n} + 1) \times C_{o u t}

Where:

Example

Given:

Image size = $6 \times 6$
Input channels = $3$
Filters: $2$ kernels of size $3 \times 3$ , each applied across all input channels

Parameters

\begin{aligned} (Given) & k_{w} & = 3, k_{h} = 3, C_{i n} = 3, C_{o u t} = 2 \\ (Formula) & Parameters & = (k_{w} \times k_{h} \times C_{i n} + 1) \times C_{o u t} \\ (Result) & = (3 \times 3 \times 3 + 1) \times 2 = 56 \end{aligned}

Output Features

\begin{aligned} (Given) & n_{i n} & = 6, p = 0, s = 1, k = 3 \\ (Formula) & n_{o u t} & = ⌊ \frac{n_{i n} + 2 p - k}{s} ⌋ + 1 \\ (Result) & = ⌊ \frac{6 + 2 \times 0 - 3}{1} ⌋ + 1 = 4 \\ (Output) & Features & = 4 \times 4 per filter \end{aligned}

Layer	Operation	Formula Inputs ( $n_{i n }, k, p, s$ )	Output Size (nout)	Output Volume	Channels $c_{i n}, c_{o u t}$	Parameters Calculation	Weights	Biases	Total Parameters
0	Input Image			227x227x3			0	0	0
1	CONV1	$n_{i n}$ =227, $k$ =11, $p$ =0, $s$ =4	55	55 × 55 × 96	$c_{i n} = 3$ $c_{o u t} = 96$	(11 × 11 × 3 + 1) × 96	34,848	96	34,944
2	POOL1	$n_{i n}$ =55, $k$ =3, $p$ =0, $s$ =2	27	27 × 27 × 96		0 (No learnable weights)	0	0	0
3	CONV2	$n_{i n}$ =27, $k$ =5, $p$ =2, $s$ =1	27	27 × 27 × 256	$c_{i n} = 96$ $c_{o u t} = 256$	(5 × 5 × 96 + 1) × 256	614,400	256	614,656
4	POOL2	$n_{i n}$ =27, $k$ =3, $p$ =0, $s$ =2	13	13 × 13 × 256		0 (No learnable weights)	0	0	0
5	CONV3	$n_{i n}$ =13, $k$ =3, $p$ =1, $s$ =1	13	13 × 13 × 384	$c_{i n} = 256$ , $c_{o u t} = 384$	(3 × 3 × 256 + 1) × 384	884,736	384	885,120
6	CONV4	$n_{i n}$ =13, $k$ =3, $p$ =1, $s$ =1	13	13 × 13 × 384	$c_{i n} = 384$ , $c_{o u t} = 384$	(3 × 3 × 384 + 1) × 384	1,327,104	384	1,327,488
7	CONV5	$n_{i n}$ =13, $k$ =3, $p$ =1, $s$ =1	13	13 × 13 × 256	$c_{i n} = 384$ , $c_{o u t} = 256$	(3 × 3 × 384 + 1) × 256	884,736	256	884,992
8	POOL3	$n_{i n}$ =13, $k$ =3, $p$ =0, $s$ =2	6	6 × 6 × 256		0 (No learnable weights)	0	0	0
9	FC1	Flattened Input: 9,216	-	4096 × 1		(9216 + 1) × 4096	37,748,736	4,096	37,752,832
10	FC2	Input: 4096	-	4096 × 1		(4096 + 1) × 4096	16,777,216	4,096	16,781,312
11	Softmax	Input: 4096	-	1000 × 1		(4096 + 1) × 1000	4,096,000	1000	4,097,000
12	Output			1000 × 1			0	0	0