Train Dogs and Cats classification using PyTorch from scratch

This repository shows Dogs and Cats classification using PyTorch from scratch. Here I use convnet for dogs vs. cats classification. I shwos you how to write a data a data loder from scratch and create a model. There are two different implementation for data loader. The first is from scratch and the second is using prebuilt "datasets.ImageFolder".

A keras implementation for this is given in the Book by "Deep learning with Python" by Francois Chollet.

Crete Virtual Environment

Create a virtiual environment.

conda create conda create --name ml_env_1.11
conda activate ml_env_1.11

Install PyTorch

conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch

Install matplotlib

conda install -c conda-forge matplotlib

Data Prepration

Download data from https://www.kaggle.com/c/dogs-vs-cats/data and uncompress it. Then execute data_prep.py to prepare train, test data directory with 2000 images of dogs and cats.

python data_prep.py

Training

To train the model on extracted data execute the following command.

python train.py

Results