Introduction

This blog will report my progress on the project of the IFT6266 course taught by Professor Aaron Courville at the University of Montréal.

The project consists in training a neural network model to classify images of dogs versus cats. The training/testing datasets come from the Kaggle competition https://www.kaggle.com/c/dogs-vs-cats who was won last year by Pierre Sermanet with a testing accuracy score of nearly 99% (he used extra data).

The whole dataset consists of 25 000 images, 12.5K of dogs and 12.5K of cats. So we won’t have any problems of class imbalance. In our course project, we will use 20K images for the training, 2.5K for the validation and 2.5K for the testing. The first objective is to obtain at least 80% of accuracy on the testing data.