Momofil31 / duce-nonduce-api

Meme API with a PyTorch deep learning model which classifies if an image provided is Duce or Non Duce

Duce-Non Duce API

Inspired by the great TV show game "Duce/Non Duce" invented by the italian comedian Valerio Lundini, this API classifies a provided image as Duce or Non Duce.

It uses a Binary Classifier built with PyTorch (Resnet18) and trained on a dataset composed of 237 images of Benito Mussolini (aka The Duce) and other common objects, animals, people and landscapes. Here the colab notebook used to train the model.

How to use (under construction)

The API receives and HTTP POST request and respond with a JSON object in a restful fashion.

This is the API URL: duce-non-duce-api.herokuapp.com

The request should be of type form-data and contain a field named file with the image you want to classify. The image should be .jpg, .png or .jpeg.

The response is a JSON string as the following format:

{ "prediction": "Duce" } or { "prediction": "Non Duce" }

If an error occurs the response is as the following:

{ "error": "prediction failed" }

Possible error codes are:

• prediction failed
• format not supported

How to run on your machine (under construction)

To run locally:

1. Create and activate a python virtual environment

   python3 -m venv venv

   source venv/bin/activate (On macOS & Linux)

   venv\bin\activate.bat (On windows)

2. Install required modules via pip

   pip3 install -r requirements.txt

3. Run the app on localhost

   cd app flask run