Cat Dog Classifier API

Contains code to query 3 models trained on the Cats VS Dogs dataset to classify between images of Cats and Dogs. The types of models are : ResNet50 (categorical), CNN (binary) and ANN (binary)

Installation

Using docker image

Pull the docker image using docker pull b1smuth/cat-dog-classifier-api:1.0.0.

To run the image using docker run -p 8000:8000 b1smuth/cat-dog-classifier-api:1.0.0 to expose port 8000 using port 8000 of the machine.

Using source code

First clone the repository with all the submodules

git clone --recurse-submodules https://github.com/bismuth01/cat-dog-classifier-api.git

This might take a while since the .h5 files are large. In the root directory of the repository, install all necessary pip packages by using the command pip install -r requirements.txt

Then to start the API server fastapi dev main.py

How to use

The API runs on port 8000 by default. To check if it's running, try a GET request on the /status endpoint.

Send a your image at /predict_image endpoint with key value file to get the prediction response.

Sample response

{
    "ResNet":{
        "prediction":{
            "class":"Cat",
            "confidence":0.9368
            }
        },
    "CNN":{
        "prediction":{
            "class":"Cat",
            "confidence"::0.9964
            }
        },
    "ANN":{
        "prediction":{
            "class":"Dog",
            "confidence":0.5294
            }
        }
}

How it works ?

Any image uploaded is resized to 128 x 128 pixels, which is the shape that the models take. For the ANN, it is also converted to greyscale since it takes only 1 color channel.

The accuracy of models in descending order is ResNet50, CNN, ANN.