Melanoma detection API

🔴 NOT FOR MEDICAL USE

Serves a deep learning based melanoma detection model using a REST API endpoint.

Local development

To use this code for local development, first clone the repo using :

$ git clone https://github.com/hasibzunair/adversarial-lesions-rest-api-demo.git

Then navigate to the project folder and install the requirements using (make sure Python version is 3.6):

$ cd adversarial-lesions-rest-api-demo
$ pip install -r requirements.txt
$ pip install -i https://test.pypi.org/simple/ melanet

Now, you're setup!

Usage

To launch the API server run:

$ python server.py

A fastapi app will run on your local machine. See Swagger UI at http://127.0.0.1:8000/docs for more info. To interact with it, open a new terminal and just send a curl request like this (make sure you are in the project folder):

$ curl -X POST -F image=@test.jpg "http://127.0.0.1:8000/api/predict"

Using the test.jpg image, the JSON response result should look like this, with labels and the probability values for the given image:

{
  "success": true,
  "predictions": [
    {
      "label": "Non-melanoma",
      "probability": 0.9180707335472107
    },
    {
      "label": "Melanoma",
      "probability": 0.08192921429872513
    }
  ]
}

Or, submit a request using Python like this:

$ python request.py

The result should look like this:

1. Non-melanoma: 0.9181
2. Melanoma: 0.0819

To try it with any of your own images(*.jpg,*.jpeg,*.png), set path to your image YOUR_IMG_PATH and run:

$ curl -X POST -F image=@YOUR_IMG_PATH "http://127.0.0.1:8000/api/predict"

Relevant materials

• Training code for the detection model: https://github.com/hasibzunair/adversarial-lesions
• Deployed streamlit demo application: https://github.com/hasibzunair/melanoma-detection-demo

License

MIT