Deploy an Image Classification Model Using Flask


Get an overview of PyTorch and Flask
Discover to construct an image classification design in PyTorch
Discover how to deploy the design using Flask.


Many of the images are identified by the image category models deployed by Instagram. And, there is also a community-based feedback loop. This is one of the most crucial use cases of the Image Classification. In this article, we will release an image category model to find the category of the images.

Image Classification is a critical pillar when it concerns the healthy functioning of Social Media. Categorizing material on the basis of specific tags are in lieu of various laws and guidelines. It becomes important so regarding conceal content from a specific set of audiences.

I regularly come across posts with a “Sensitive Content” on some of the images while scrolling on my Instagram feed. Any image on a humanitarian crisis, terrorism, or violence is normally categorized as Sensitive Content.

Table of Contents

What is Model Deployment?
Introduction to PyTorch
What is Flask?
Setting Up Flask and PyTorch on your Machine
Understanding the Problem Statement
Setup the Pre-Trained Image Classification Model
Construct an Image Scraper
Develop the Webpage
Setup the Flask Project
Working of the Deployed Model

What is Model Deployment?


Easy to utilize API– The PyTorch API is as simple as python can be.

Keep in mind: If you are brand-new to PyTorch, I would advise you to go through below resources:.

Python support– PyTorch smoothly integrates with the python information science stack.


Dynamic calculation charts– PyTorch supplies a framework for us to build computational charts as we go, and even change them throughout the runtime. This is valuable for circumstances where we dont know just how much memory is going to be required for developing a neural network.

In additional sections, we will utilize a pre-trained design to identify the classification of the image utilizing PyTorch. Next, we will be using Flask for design implementation. In the next area, we will briefly talk about Flask.

Introduction to PyTorch.

PyTorch is a python based library that supplies versatility as a deep learning development platform. The workflow of PyTorch is as close as you can get to pythons clinical computing library– NumPy

In a common machine learning and deep knowing job, we typically start with specifying the issue declaration followed by information collection and preparation, and model structure?

Enroll in the complimentary course of PyTorch: Introduction to PyTorch for Deep Learning.

PyTorch is being commonly used for building deep learning designs. Here are some essential benefits of PyTorch–.

A fantastic tutorial: An Introduction to PyTorch– A Simple yet Powerful Deep Learning Library.

As soon as we have successfully built and trained the model, we desire it to be offered for the end-users. Hence we will have to “release” the design so that the end-users can use it. Model Deployment is one of the later stages of any artificial intelligence or deep knowing project.

In this article, we will build a category design in PyTorch and then find out how to deploy the same utilizing Flask. Before we enter into the details, let us have a fast introduction to PyTorch

What is Flask?

Construct an Image Scraper.

Now, our model is all set to anticipate the classes of the image. Lets start with developing the image scraper

Produce an Image Scraper: We will develop a web scraper utilizing demands and the BeautifulSoup library. It will download all the images from a URL and store it so that we can make forecasts on it.


Now, we will specify a function get_images. It will initially produce the directory using the get_path function and then send out a request for the source code. From the source code, we will draw out sources by utilizing “img” tag.

! pip install torch torchvision.

Classify images and send out results: Once we get the query from the user, we will use the design to anticipate classes of the images and send out the outcomes back to the user.

Here is a representation of the actions we just saw:.

Here is the workflow for the end-to-end design-.

Setting up Flask is simple and uncomplicated. Here, I am assuming you currently have Python 3 and pip installed. To set up Flask, you need to run the following command:.

get_category( image_path= static/sample _ 2. jpeg)
## [ n11939491, daisy]

Setting Up Flask and PyTorch on your Machine.

Comprehending the Problem Statement.

Next, we will specify a function to get the classification of the image. For this, we will pass the course of the image as the only parameter. At first, it will read the image and open in the binary type and after that change it. It will pass the transformed image to the design to get the forecasted class. It will utilize the mapping and return the class name.

We will start by importing some of the needed libraries. For each of the URLs that we will scrape a new directory will be created to save the images. We will create a function get_path that will return you the path of the folder produced for that URL.

And, here the task is to scrape all images from the URL. For each of the images, we will forecast the classification or class of the image using an image classification model and render the images with classifications on the website.

Thats it! Now let us use up an issue declaration and construct a design.

After this, We will pick only the images with jpeg format. You can also add png format images. I have filtered them out as most of the png format images are logos. In the end, start the counter and conserve images with counter names in the specified directory site.

Lets attempt this function on a few images:.

Model Building: We will utilize a pre-trained model Densenet 121 to predict the image class. It is offered in the torchvision library in PyTorch. Here, our focus will not be on building an extremely precise classification design from scratch however to see how to deploy that design and use it with a web user interface.

Style Webpage Template: Also we will design a user interface where the user can submit a URL and also get the outcomes when determined.

You can download the complete code and dataset here.

Here is a sample of the mapping:.

We will use a pre-trained model Densenet 121 to classify the images. If you wish to construct an Image Classification model I would highly suggest you to go through this post: Build your First Image Classification Model in simply 10 Minutes!

. Flask offers a range of choices for developing web applications and it gives us the essential tools and libraries that permit us to build a web application

Next, we need to set up the PyTorch. You are not needed to have the GPU to run the code provided in this post.

Now, the pre-trained model returns the index of the predicted class id. PyTorch has provided mapping for the exact same so that we can see the name of the anticipated class. You can download the mapping here. It has 1000 various categories.

Setup the Pre-Trained Image Classification Model.

sudo apt-get set up python3-flask.

Lets begin by importing some of the needed libraries and get the densenet121 model from the torchvision library. Make sure to include the specification pretrained as True.

We will use the BeautifulSoup library to download the images. You are free to use any other library or an API that will provide you the images.

Establishing the Project WorkFlow.

Lets discuss all the needed elements of the tasks:.

get_category( image_path= static/sample _ 1. jpeg)
## [ n02089973, English_foxhound]


Flask is a web application framework written in Python. It has several modules that make it much easier for a web designer to write applications without needing to stress over the details like protocol management, thread management, and so on

Now, we will define a function to transform the image. It will produce a transform pipeline and transform the image as needed.

Lets try the scraper that we have simply produced!

Develop the Webpage.

” home.html” is the default one which will have a text box in which a user can type the URL.
” image_class. html” will help us to render the images with their categories.

We will create two web pages one is “home.html” and another one is “image_class. html”.

Now, a new directory site is produced and see how it looks like. We have all the images downloaded at a single place.

Keep in mind: It is recommended to use this Image Scraper as per the knowing purpose just. Always follow the robots.txt file of the target website which is likewise referred to as the robotic exemption protocol. This tells web robotics which pages not to crawl.


1. home.html.


We need to add the kind tag in the home.html file to collect the information in the search container. In the kind tag, we will pass the technique post and name as “search”

By doing this, our backend code would have the ability to know that we have received some data with the name “search”. At the backend, we require to process that information and send it.

2. image_class. html.

We have actually done the following tasks included in our project:.

Here, is code to perform this:.

Flask application will first render the home.html file and whenever someone sends a demand for the image category, Flask will spot a post technique and call the get_image_class function.

Setup the Flask Project.

Lets have an appearance at the directory site structure.

Running a Flask Application.

And now we need to link all these files together so that we can have a working job.

Keep in mind: Make sure that you save the images in the folder name html and fixed files in templates. Flask will only try to find these names. You will get a mistake if you alter these.

Image classification.
Frequency count of all available image category.

Image Classification design that is working fine and able to classify the images.
We have built the Image scraper that will download the images and keep them.
We have actually produced the webpage to get and return the results.

The next action is to setup the Flask task to integrate these individual pieces to resolve the challenge.

While determining the outcomes another page will get rendered with the outcomes as shown listed below. And you can see that we are showing below information on the web page:.

It will send a request to download the images and keep them.
Next, It will send out the directory site path to the get_prediction. py file which will compute and return the results in the kind of a dictionary.
Lastly, It will send this dictionary to the generate_html. py, file generating the output file which will be returned to the user.

This function will work in the following steps:.

Later on, we will send this dictionary to the generate_html. py file which will develop the HTML file for us.

Most of the images are identified by the image category models deployed by Instagram. In this post, we will deploy an image category design to detect the classification of the images.

Lets check out the working of the released model.

To start with, produce a things of the Flask class that will take the name of the existing module __ name __ as an argument. The route function will inform the Flask application which URL to render next on the web page.

Once the above actions are done, we are all set to serve the user with the results. We will call the success function which will then render the image_class. html file.

Design Building: We will utilize a pre-trained model Densenet 121 to anticipate the image class. We dived into comprehending numerous steps involved in the procedure of creating an image classification model using PyTorch and deploying it with Flask.

The model was released on the localhost. We can likewise deploy it on Cloud Services like Google Cloud, Amazon, etc. We will cover this likewise in the upcoming article.

Till now, we have taken prediction for each image separately. Now, we will fix this by customizing get_category function with new parameters. We will pass the directory site course which will include numerous image files.

Open the web browser and go to localhost:5000 and you will see that the default web page is rendered there. Now, type any URL in the text box and press the search button. It might take 20-30 seconds depending upon the variety of images because URL and the Internet speed.

Now, we will run the file get_class. py and the flask server will start on localhost:5000.

You can download the total code and dataset here..

Now, we will define another function get_prediction which will utilize the get_category function and will return the dictionary where the secrets will be the image course and the worths will be the image class.

Related Articles.

In this post, I discussed, in quick, the ideas of model release, Pytorch, and Flask. Then we dived into comprehending different actions associated with the process of creating an image category model using PyTorch and releasing it with Flask. I hope this helps you in building and releasing your image classification design.

For each of the images, we will forecast the category or class of the image using an image category design and render the images with categories on the website.

You can also read this post on our Mobile APP.

Connect on the remark section in case of any doubts. I will enjoy to assist.

Working of the Deployed Model.

Get Prediction for all pictures of Source URL.

Now, all the code files are ready and we simply need to link these with the master file.

End Notes.