Cat_Or_Dog

Abstract

This is Convolutional Neural Network classification to classify cats and dogs built with keras in Tensorflow. There are a total of 10,000 images of cats and dogs considered, out of which 8000 are used for training and 2000 are used for testing. The images are rescaled to 150 X 150 sized images. The model is trained with 30 epochs and a final accuracy of 94.44% is achieved for Epoch 30/30.

Improvements

Initially the model was trained with image rescaling of 64 X 64 and for 25 epochs the final accuracy obtained was 91%, however by increasing the scaling of the image and the number of epochs the accuracy increased by 4.44%

Deployed

The model is deployed with the help of flask API in python, with a simple UI. You can choose an image file and upload the same into the website and click on "Predict" button the website replies with a message indicating weather the image is cat or dog. However this is a bi-class classification model it classifies every image to either cat or dog and nothing else.

References

1. Image data loading : https://keras.io/api/data_loading/image/
2. Save and load keras modules : https://www.tensorflow.org/guide/keras/save_and_serialize
3. Deployment of machine Learning model : https://www.youtube.com/watch?v=0nr6TPKlrN0&t=9s
4. CSS Bootstrap : https://getbootstrap.com/docs/3.4/css/

Disclaimer