AI Image classifiers on Django server with REST API

Table of Contents

• Introduction
• Description
• Features
• Architecture
• Usage
• Results
• Issue
• Others
• License

Introduction

A Keras deep learning image classifiers on Django server with REST API. It can help you quickly deploy and apply ML models.

Description

Machine Learning (ML) models are typically trained and test on benchmark data sets. However, such data sets may not represent the data in real applications. These ML models should be updated after tested on real data sets. Deployment of the models after each update may be time consuming. Therefore, automated deployment is needed to save time and effort. In this proposal, we describe our solution to integrate trained ML models into any application of interest. We use Django web-based platform to build REST API for model deployment. Jenkins is used for continuous integration and automated deployment of ML models into Django servers.

Features

• Add Account management funtion
• Add Login & Signup
• Add Django Rest Framework
• Optimize front-end interface, adapt to mobile and PC interface
• Design front-end and back-end interactive interfaces
• Add feature: Face comparison
• Add feature: Bank card identification
• Add feature: Gesture identification
• Add feature: Image classify using following model：
  • ResNet50
  • Xception
  • MobileNet, MobileNetV2
  • InceptionV3, InceptionResNetV2
  • DenseNet121, DenseNet169, DenseNet201
  • VGG16, VGG19
  • NASNetMobile, NASNetLarge
• Run each part of functions seperately in local and cloud server
• Run all the functions in local server and cloud server
• Design and using machine learning CI/CD Tools such as Git, Jenkins, Nginx, uwgsi for deployment.

Architicture

Usage

I assume you already have your own local virtual environment.

git clone https://github.com/nature1995/image-classify-django-server.git

pip install -r requirements.txt
python manage.py makemigrations
python manage.py migrate
python manage.py collectstatic
python manage.py runserver 0.0.0.0:8000

Access the web page though this link: http://127.0.0.1:8000/

Name	Description
Face comparison	Extract and analyze facial features in pictures
Bank card identification	Identify the field information above the bank card
Gesture identification	Detect and return the meaning of the gesture in the picture.
Image classify	Detects objects in the image, returns the detected object name, and the corresponding confidence.

Docker

You can use dockerfile to build this project easily.

# Follow any instruction to install the docker-ce, docker for OS X or PC in your device
docker build -t ai_image_classifiers .
docker run -it --rm -p 8000:8000 --name ai_image_classifiers_app ai_image_classifiers:latest

Result

Name	Input Size	API address
ResNet50	224x224	http://127.0.0.1:8000/predict/api/
Xception	299x299	http://127.0.0.1:8000/predict_Xception/api/
MobileNet	224x224	http://127.0.0.1:8000/predict_MobileNet/api/
MobileNetV2	224x224	http://127.0.0.1:8000/predict_MobileNetV2/api/
InceptionV3	299x299	http://127.0.0.1:8000/predict_InceptionV3/api/
InceptionResNetV2	224x224	http://127.0.0.1:8000/predict_InceptionResNetV2/api/
DenseNet121	224x224	http://127.0.0.1:8000/predict_DenseNet121/api/
DenseNet169	224x224	http://127.0.0.1:8000/predict_DenseNet169/api/
DenseNet201	224x224	http://127.0.0.1:8000/predict_DenseNet201/api/
VGG16	224x224	http://127.0.0.1:8000/predict_VGG16/api/
VGG19	224x224	http://127.0.0.1:8000/predict_VGG19/api/
NASNetLarge	331x331	http://127.0.0.1:8000/predict_NASNetLarge/api/
NASNetMobile	224x224	http://127.0.0.1:8000/predict_NASNetMobile/api/

Input

Parameter	Type	Description
image	file	Image file that you want to classify.
top	text (optional, default=6)	Return top-k categories of the results. Must me string in integer format.

Note: You can not send a very large size image.

Result

Parameter	Type	Description
success	bool	Whether classification was sucessfuly or not
predictions	label, float	pair of label and it's probability

Accuracy for individual models

Model	Accuracy	Top-6 Accuracy
Xception	0.780	0.955
VGG16	0.722	0.914
VGG19	0.723	0.910
ResNet50	0.748	0.931
InceptionV3	0.789	0.948
InceptionResNetV2	0.813	0.963
MobileNet	0.714	0.904
MobileNetV2	0.723	0.912
DenseNet121	0.761	0.934
DenseNet169	0.772	0.952
DenseNet201	0.783	0.945
NASNetMobile	0.752	0.920
NASNetLarge	0.837	0.959

Example

Using Postman to test the API:
POST http://127.0.0.1:8000/predict_VGG19/api/

Result

{
    "success": true,
    "predictions": [
        {
            "label": "red_fox",
            "probability": 0.8969062566757202
        },
        {
            "label": "kit_fox",
            "probability": 0.08841043710708618
        },
        {
            "label": "grey_fox",
            "probability": 0.012036639265716076
        },
        {
            "label": "Arctic_fox",
            "probability": 0.0022438077721744776
        },
        {
            "label": "coyote",
            "probability": 0.0002566342300269753
        },
        {
            "label": "white_wolf",
            "probability": 0.00005685776341124438
        }
    ]
}

Compatibility

Support to Django>=2.1.5 or Django==3.0.0 and Python 3.5, 3.6, 3.7.

Issue

If you have questions or issues, please feel free to tell us.

Contribution

Welcome to make pull request. If you have a related project/component/tool, add it with a pull request to add it!

Others

Admin Account

python manage.py createsuperuser

username: ranxiaolang
email: YOUR EMAIL  
password: ranxiaolang  

Access the web page though this link: http://127.0.0.1:8000/admin

Django Restframework

Access the web page though this link: http://127.0.0.1:8000/iot/

Contributors

@nature1995

@sergulaydore

License

MIT ©