Car Damage Detection, Masking and Area Percentage Calculation

The following code adapts from the Mask_RCNN implementation from the Facebook AI Research Team (FAIR).

• Main source repository: https://github.com/matterport/Mask_RCNN

Run inference on images

To run on Google Colab with zero configuration, copy this notebook to your own google drive.

To run locally follow these steps:

Prepare a python 3.7 environment.

Open a terminal and execute the following lines:

git clone https://github.com/Rajat-Roy/car_damage.git
cd car_damage
pip install -r requirements.txt
wget https://github.com/Rajat-Roy/car_damage/releases/download/v1.0/mask_rcnn_car_damage_0030.h5
wget https://github.com/Rajat-Roy/car_damage/releases/download/v1.0/dataset.zip

Extract the dataset.zip archive

If jupyter notebook is not installed already:

pip install jupyter

Start the notebook server:

jupyter notebook

In the jupyter app directory listing click and run the car_damage_demo.ipynb file.

Change the file paths in the notebook accordingly.

Execute the code cells.