Car Exterior Body Parts Localization

A deep learning-based project that aims to localize 14 exterior car body parts in case of no or partial damage.
The project was implemented using both Yolov5 and Yolov8.

Dataset Sources

1. DSMLR Body Parts Segmentation Dataset Github.
2. CompCars Dataset.
3. Stanford Cars Dataset.
4. Used cars Auction Nation live auctions.

Project Approach

1. The data collected from sources 2-4 were manually annotated using Microsoft VOTT.
2. Annotations from source 1 were edited and the labels out of intreset were removed.
3. Data was splitted over train, validation, and test.
4. Transfer learning was deployed as Yolov5 and Yolov8 state-of-the-art object detection CNNs were tuned and trained.

Repo Content

1. How to use guide : setup.md
2. Source code as jupyter notebook : BodyParts.ipynb
3. Detailed compare between Yolov8 and Yolov5 results : results_discussion.md

Results

The model was capable of detecting the body parts even the partially damaged.

bodyparts.mp4

Contributors

Doha ElHady - Mohamed Elahl - Hassan Mohamed - Karim Youssef

Dataset

The intial dataset is availble at DSMLR Body Parts Segmentation Dataset.

The complete dataset manipulated and manually annotated is available upon request.

Request complete dataset!