A Pytorch project for garbage classification using the EfficientNet-B6 model to achieve a 95.78% accuracy on the test set.
Important
This project is a new version of the original project, which can be found here but with a significantly improved training process + code and a different dataset.
This project is not hard coded for this specific dataset, so it can be used for any image classification task and it has all the necessary tools to train a model from scratch and more. (I will make a pytorch classification template soon)
The dataset used for this project is the Garbage Classification from Kaggle. It contains images of garbage, divided into six categories.
├───Database
│ └───Data # Put all the folders with images here
# Example ⬎
│ ├───battery
│ ├───biological
│ ├───brown-glass
│ ...
│ └───white-glass
I used the EfficientNet-B6 model for this project. EfficientNet-B6 is a convolutional neural network that is pretrained on the ImageNet dataset. It is known for its efficiency and high performance on a variety of image classification tasks. Original paper
To run the code in this repository, you will need to install the required libraries. You can do this by running the following command:
pip install -r requirements.txtWarning
The requirements are auto generated by pipreqs and may not contain all the necessary dependencies. like hidden ones like Tensorboard.
The main code for this project is in a Jupyter notebook named Main.ipynb. To run the notebook, use the following command:
jupyter notebook Main.ipynb| Metric | Value |
|---|---|
| Loss | 0.0330466 |
| F1 Score (macro) | 0.95472 |
| Precision (macro) | 0.952111 |
| Recall (macro) | 0.957959 |
| AUROC | 0.993324 |
| Accuracy | 0.957839 |
| Cohen's Kappa | 0.948292 |
| Matthews Correlation Coefficient | 0.948374 |
Copyright (c) 2024 Aydin Hamedi This software is released under the MIT License. https://opensource.org/licenses/MIT