ENSC 813 Deep Learning Sytems in Engineering - Term Project
This project implements ConvNets for binary and multi-class classification of car images in the The Car Connection dataset.
The The Car Connection dataset was scraped by Mr. Nicolas Gervais. The provision of this dataset for free is gratefully acknowledged. The source repo for this dataset is Predicting car price from scraped data.
- Hans Dhondea - This repo - AshivDhondea
The project report and its source files can be found in this repo.
A brief user manual for this code can be found in the user manual directory of the report repo.
Due to GitHub's file size restrictions, most models cannot be committed to this repo. The model and weights files created by main_22_multiclass_00.py can be found in the output_files directory.
This project was created using a custom conda environment called tfgpu.
To re-create this environment, the following command may be run in the Anaconda Powershell Prompt: conda create -n tfgpu --ensc813-tfgpu-package-list.txt
. More info in the requirements directory.
Python packages used are listed in the file ensc813-tfgpu-package-list.txt which can be obtained by running the command conda list --export > ensc813-tfgpu-package-list.txt
.
For the binary classification task, we made use of the following ConvNet architectures
We modified the binary classification ConvNets to accommodate multi-class classification problems. We ensembled our ConvNets to obtain an improved multi-class classifier. The following confusion matrix summarizes its classification performance.
If you use this work, cite it as
@misc{Hans2020DL,
author = {Dhondea, A.R.},
title = {Classifying car images in the TCC Dataset},
year = {2020},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/AshivDhondea/ENSC813_Project}}
}
This project is licensed under the MIT License - see the LICENSE.md file for details