Simple Image Classifier

This repository contains the my adapted approach to training a simple image classification model based on the popular Kaggle notebook by Rob Mulla (https://www.kaggle.com/code/robikscube/train-your-first-pytorch-model-card-classifier)

Getting Started

• Clone this repo to your local machine.

git clone https://github.com/MayorDobe/Simple_Image_Classifier.git

Set up enviroment.

• conda

conda create -n simple_image_classifier python=3.13
conda activate simple_image_classifier

• venv

python -m venv /path/to/new/virtual/environment
source /path/to/new/virtual/enviroment/bin/activate

• Install required packages.
• You may need to install pytorch independently. https://pytorch.org/

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126
pip install -r requirements.txt

• Generate a Kaggle api key to download our dataset

Authentication https://www.kaggle.com/docs/api

In order to use the Kaggle’s public API, you must first authenticate using an API token. Go to the 'Account' tab of your user profile and select 'Create New Token'. This will trigger the download of kaggle.json, a file containing your API credentials.

• Once we have our api key in ~/.kaggle/kaggle.json we can pull down our dataset.

cd ~/Simple_Image_Classifier
mkdir dataset && cd dataset
kaggle datasets download gpiosenka/cards-image-datasetclassification --unzip

Usage.

The project contains two scripts one for training our model with a limited number of parameters that can be changed located in the config dictionary of training_script.py

config = {
    "model": "efficientnet_b0",
    "optimizer": optim.Adam,
    "criterion": nn.CrossEntropyLoss(),
    "batch_size": 32,
    "num_epochs": 6,
    "learning_rate": 0.001,
    "device": "cuda" if torch.cuda.is_available() else "cpu",
}

To train the model we execute our script. after completion a graph will plot our loss.

python training_script.py

To evaluate our model we run our evauation script. This will give us our models accuracy rating along with a number of set examples showing our models probabilities.

python evaluation_script.py

TODO

• Add functionality to allow command line arguments to be passed.
• Create a config.json file for easier manipulation of hyper-params
• Add functionality to allow more than a single model to be trained and saved.
• Add functionality to display better metrics of our model.
• learn
• fail
• succeed
• repeat