In this repository, I build a neural network which takes in a picture of a person and returns whether that person is wearing glasses or not. It is based on an old project I did Glasses-Nerual-Network except this time it is a proper machine learning code-base with multiple functionalities for training which I learnt from the Full Stack Deep Learning Course.
The code-base makes heavy use of Pytorch and Pytorch-Lightning.
Using conda, create an environment using:
conda create --name myenv python=3.10 pytorch torchvision torchaudio cudatoolkit=11.8 -c pytorch -c nvidiaand then install the packages in requirements.txt using:
pip install -r requirements.txt
In the future, I plan to setup an environment.yaml file to make setup easier similar to what is done in the Full Stack Deep Learning Course Repo
Finally run the following path to set the Python Paths:
echo "export PYTHONPATH=.:$PYTHONPATH" >> ~/.bashrcUnder glasses_detector/data you will find the code for all the data modules used to download and setup our data. Specifically, glasses.py contains the data module for the glasses detector model.
Under glasses_detector/data/utils.py you will find code to set up a PyTorch Dataset class, split the dataset and download the data from a URL.
Under glasses_detector/models you will find some CNN models you can train in the code-base.
Under 'glasses_detector/lit_modelsyou will find aLightningModule` class which specifies training and validation steps, along with how to log metrics and setup optimizers.
Under glasses_detector/metadata you will find metadata for the project such as URLS, image sizes, URLs, etc.
You can find the original dataset in this repo. I also want to credit the Full Stack Deep Learning Course as I used most of the code from there as a guideline to how I should structure this repository.