# Depthwise Separable Convolutions using Tensorflow in Python

In this tutorial, we are going to learn about Depthwise Separable Convolution using the Tensorflow library in Python.

Let’s first understand what we meant by Depthwise Convolution:

## What is Depth-wise Separable Convolution?

Let’s first understand what is convolution. Convolution is a process of applying a kernel over volume, where a weighted sum of pixels with the weight as values of the kernels in image processing.

Now, Let’s understand Depth-wise Convolution. A Depth-wise Convolution is a type of convolution in which we perform convolution along only one spatial dimension of the image. It differs from the normal convolution model because the normal convolution model is applied across all spatial dimensions at each step.

The Depth-wise Separable Convolution deals not only with the spatial dimensions but also with the depth dimension. It is based on the idea that a filter’s depth and spatial dimension can be separated.

## Explanation of Code

Let’s first install the required Python libraries:

`pip install tensorflow`
`pip install keras`

Now let’s import all the required libraries:

```import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
from tensorflow.keras import models```

Depth-wise Separable Convolution layer using in TensorFlow:

```input_depth=tf.keras.layers.Input((110,124,5))
layers1=tf.keras.layers.SeparableConv2D(17,(3,3),activation='relu')(input_depth)
layers1=tf.keras.layers.Flatten()(layers1)
layers1=tf.keras.layers.Dense(1)(layers1)

model_depth=tf.keras.models.Model(input_depth,layers1)
model_depth.summary()```

Output:

Depth-wise convolution also includes parameter  called depth_multiplier :

```input_depth=tf.keras.layers.Input((110,124,5))
layers1=tf.keras.layers.SeparableConv2D(17,(3,3),activation='relu',depth_multiplier=3)(input_depth)
layers1=tf.keras.layers.Flatten()(layers1)
layers1=tf.keras.layers.Dense(1)(layers1)

model_depth=tf.keras.models.Model(input_depth,layers1)
model_depth.summary()```

Output:

In this way, you can practically use the Depth-wise Separable Convolution layer using TensorFlow in Python.