Semantic Segmentation Web and API Project

Web application and REST Api to serve DeepLabV3+ model to perform semantic segmentation on the input image.

Tech stack

Backend - Django, Django Rest Framework
Frontend - HTML, CSS, JavaScript, Bootstrap, Jinja Templating Engine
To test apis - Postman
For virtual environment- pyenv

Working

Web: Run at /predict, Upload Image and it will segment the image and then show both the images.
API: Upload the image with a POST request to the server at /predict-api with an image as 'img' field and it will respond with a path to the location of input & output images.
Output Format of API:

{ "input_image_url":"url_wrt_server", "output_image_url":"url_wrt_server" }

APIs

/predict --> Return webpage with functionality to upload an image and get inferred image.
/about --> Returns about page
/predict-api --> Takes post requests with an image field as 'img' and returns path to input & inferred image.

Screenshots

Website front page

After prediction

API response in postman

About page

Running django server using vscode

Classes it is able to predict

Tested on

Python 3.6.8
Django 3.0.6
Django rest framework 3.11.0
django-cors-headers 3.2.1
tensorflow 1.15.2
OS - Ubuntu 20 LTS

Directories

/frontend --> To test apis
/media/images --> Stores Input Images
/predict-with-deeplabv3 --> contains Deeplabv3 model and scripts
/prediction --> App with prediction logic
/sample-images --> Sample images to test
/semantic-segmentation-api --> Main folder, contains settings
setup-env.sh --> Bash script to setup env required for the model

Steps to run locally (On Linux based distro)

Install pyenv
Open terminal from the root location of this project
Run command: ./setup-env.sh
Run: pyenv local venvSSA
Run: python manage.py migrate
Run: python manage.py migrate --run-syncdb
Run: python manage.py runserver

References

Credits

Special thanks to original the author of DeeplabV3+ and its implementation
Mr Himanshu Mittal for guidance on major project

Note

Size of model is large in terms of disk space, making size of project large.
It would be better to use docker to run the model.
It would be better to use a webserver like nginx to serve inferred images.
Not for production (It is a simple College Project)