Home

CNN

Name	Input Dimension	Filter Dimension	Output Dimension	Number of Filters	Link
2d filter	(224,224)	(3,3)	(224,224)	1	Sobel
2d conv	(224,224,3)	(3,3,3)	(224,224)	64	VGG
	(224,224,64)	(3,3,64)	(224,224)	128	VGG
	(112,112,128)	(3,3,128)	(112,112)	256	VGG
1x1 conv	(224,224,64)	(1,1,64)	(224,224)	32	VGG
3d conv	(224,224,3,16)	(3,3,3,16)	(224,224,16)	64	C3D
MobileNets conv	1

Tensorflow

batch_size = 10

Input Image = [width, height, channels] = [224,224,3(RGB)]

Name	Input	Filter	Input Shape	Filter Shape	Output Shape
tf.nn.conv1d	3D	3D	2D	2D	1D
	[10,100,64]	[3,64,128]	[100,64]	[3,64]	[100] x 128
tf.nn.conv2d	4D	4D	3D	3D	2D
	[10,224,224,64]	[3,3,64,128]	[224,224,64]	[3,3,64]	[224,224] x 128
tf.nn.conv3d	5D	5D	4D	4D	3D
	[10,3,224,224,64]	[3,3,3,64,128]	[3,224,224,64]	[3,3,3,64]	[3,224,224] x 128

API	Input
tf.nn.conv1d	batch	in_width	in_channels
tf.nn.conv2d	batch	in_height	in_width	in_channels
tf.nn.conv3d	batch	in_depth	in_height	in_width	in_channels

API	Output
tf.nn.conv1d	batch	filter_width	out_channels
tf.nn.conv2d	batch	filter_height	filter_width	out_channels
tf.nn.conv3d	batch	filter_depth	filter_height	filter_width	out_channel

Recap	conv direction	output shape
conv1d	only x-direction	1-dim array
conv2d	x,y direction	2-dim matrix
conv3d	x,y,z dirction	3-dim volume