This repository contains Bradley Reardon, Divya Parmar and Haruna Salim's final project for George Washington University's DATS 6203:Machine Learning II course. Our project objective was to apply a deep learning framework and network to a real world problem. For our project we applied preprocessing techniques and a convolutinal neural network to classify cars by make and model.
- Code contains all of our code and data used in the project.
- Proposal contains the proposal for the project.
- Presentation contains a PowerPoint presentation of our project.
- Report contains a report of our findings from this project.
- To successfully execute the code, make sure you have following libraries installed on your python interpreter enviroment:
- sklearn
- opencv-python
- opencv-contrib-python
- tensorflow
- keras
- seaborn
- pandas
- matplotlib
-
After downloading the training images from here and the test images from here, copy the images from the car_test into the Dataset/Test folder and the images from the car_train into the Dataset/Train folder.
-
To cut data preprocessing time, you can access the already preprocessed numpy files here. Simply copy the train_data.npy and test_data.npy into the DataStorage folder in the Code folder. If you also want to reprocess the data, just execute the DataGenerator.py file again.
-
When you run the main.py file. Set either pretrained or custom to true or false or both.
-
After the preprocessing and trainingoccurs, the model will be saved to the SavedModel folder and will be used for prediction on the test set.