Unofficial PyTorch Module - Depthwise Separable Convolution

An illustration of Depthwise Separable Convolution. Credit: Depthwise Convolution Is All You Need for Learning Multiple Visual Domains.

PyTorch (unofficial) implementation of Depthwise Separable Convolution. This type of convolution is introduced by Chollet in Xception: Deep Learning With Depthwise Separable Convolutions. This package provides SeparableConv1d, SeparableConv2d, SeparableConv3d, LazySeparableConv1d, LazySeparableConv2d, and LazySeparableConv3d.

Installation

Install separableconv-torch using pip (require: Python >=3.7).

pip install separableconv-torch

Parameters

Parameter	Description	Type
in_channels	Number of channels in the input image	int
out_channels	Number of channels produced by the convolution	int
kernel_size	Size of the convolving kernel	int or tuple
stride	Stride of the convolution. Default: 1	int or tuple, optional
padding	Padding added to all four sides of the input. Default: 0	int, tuple or str, optional
padding_mode	'zeros', 'reflect', 'replicate' or 'circular'. Default: 'zeros'	string, optional
dilation	Spacing between kernel elements. Default: 1	int or tuple, optional
depth_multiplier	The number of depthwise convolution output channels for each input channel. The total number of depthwise convolution output channels will be equal to in_channels * depth_multiplier. Default: 1	int, optional
normalization_dw	depthwise convolution normalization. Default: 'bn'	str, optional
normalization_pw	pointwise convolution normalization. Default: 'bn'	str, optional
activation_dw	depthwise convolution activation. Default: torch.nn.ReLU	Callable[..., torch.nn.Module], optional
activation_pw	pointwise convolution activation. Default: torch.nn.ReLU	Callable[..., torch.nn.Module], optional
bias	If True, adds a learnable bias to the output. Default: True	bool, optional

Example Usage

For 1-dimensional case.

import torch
import separableconv.nn as nn

# set input
input = torch.randn(4, 10, 100)

# define model
m = nn.SeparableConv1d(10, 30, 3)

# process input through model
output = m(input)

For 2-dimensional case.

import torch
import separableconv.nn as nn

# set input
input = torch.randn(4, 10, 100, 100)

# define model
m = nn.SeparableConv2d(10, 30, 3)

# process input through model
output = m(input)

For 3-dimensional case.

import torch
import separableconv.nn as nn

# set input
input = torch.randn(4, 10, 100, 100, 100)

# define model
m = nn.SeparableConv3d(10, 30, 3)

# process input through model
output = m(input)

Stacked SeparableConv2d.

import torch
import separableconv.nn as nn

# set input
input = torch.randn(4, 3, 100, 100)

# define model
m = nn.Sequential(
        nn.SeparableConv2d(3, 32, 3),
        nn.SeparableConv2d(32, 64, 3),
        nn.SeparableConv2d(64, 96, 3))

# process input through model
output = m(input)

For lazy 2-dimensional case.

import torch
import separableconv.nn as nn

# set input
input = torch.randn(4, 10, 100, 100)

# define model
m = nn.LazySeparableConv2d(30, 3)

# process input through model
output = m(input)