In this project, we use Deep Learning techniques like CNNs and FCNNs to identify tire damage from a dataset of tire images. Our team has chosen this problem as an interesting application of Deep Learning in the automotive industry.
Context The dataset consists of 1028 images of tires in total.
The dataset used for this project can be accessed via the Harvard Dataverse
This dataset is split into training and testing data, which is further split into Cracked (Oxidized) and Normal Tires. It can be used for binary classification.
This repository contains the following notebooks:
- 01_explore.ipynb - Data exploration and visualization
- 02_preprocess.ipynb - Data preprocessing and preparation
- 03_Handcrafted_Model_Lenet.ipynb - Implementation of the handcrafted model (LeNet architecture)
- 04_model_hyper_search.ipynb - Hyperparameter search and comparison with the handcrafted model
- 05_model_handcrafted_second_trial.ipynb - Second iteration of the handcrafted model
- 06_tranfer_model.ipynb - Implementation of transfer learning and comparison with the best model
Each notebook contains a report section detailing the approach, steps, and findings of the corresponding part of the project.
To use the code, simply open the respective Jupyter Notebook and run the cells in the given order. Make sure to modify the data paths if necessary to match your local setup.
Feel free to submit issues or pull requests if you find any issues or improvements that could be made. Please make sure to provide a detailed description of the changes made.