fagan2888 / object-detection-api

A django web-app that provides object detection using YOLOv3 and an API.

Object-Detection-API

A web-app that provides object detection using YOLOv3 and also an API.

It's implemented using django framework and PyTorch (for YOLO model). The app uses cloudinary API for image hosting.

Dependencies

• Django
• PyTorch
• Pillow
• OpenCV

You also need to download the yolo.weights file and place it in the "weights" directory.

You can download the weights by -

    $ wget https://pjreddie.com/media/files/yolov3.weights

Usage

Firstly, you need to put your cloudinary credentials (cloud_name, api_key and api_secret) in settings.py file.

To run the server

    $ pip3 install -r requirements.txt
    $ python3 manage.py collectstatic
    $ python3 manage.py runserver

The website is hosted at - https://object-detection-yolo.herokuapp.com

The detection service is currently not working because of memory limits of Heroku.

Web API

To use the web API, you can send a POST request to - https://object-detection-yolo.herokuapp.com/detect/api

Input

You can send either of the following parameters -

Parameter	Type	Description
image	file	Image file that you want to detect.
image64	text	Image in base64 form that you want to detect. Currently supports JPEG images only

Result

Parameter	Type	Description
success	bool	Whether classification was successful or not
detect	class label, float	pair of label and its confidence
url	image URL	Link to the result image uploaded to cloudinary

Example: {"success": true, "detect": { "dog": 0.9989, "truck": 0.9999 }, 'url':"https://example.com/image.png"}
"detect" will be empty if no objects are detected.

Example

The website also shows the detection output with bounding boxes around the detected objects. There will be no box if the input doesn't contain any object.

Input

Output

No. of objects detected - 3

The result image is generated using matplotlib.

TO-DO

• Move app to Google Cloud Platform
• Create a docker image

Contribute

If you want to contribute and/or find any bug, feel free to do a pull request!