Overview

• This repository provides an overview of implementing a Mask-RCNN on a custom image dataset to detect 2 classes of damages to automobile exterior bodies - DENT and SCRATCH

Custom Dataset

• 100 images of damaged automobiles with scratches and dents were collected from google (Training-80; Validation-20)
• The regions of damages were annotated using the VIA annotation tool (https://www.robots.ox.ac.uk/~vgg/software/via/via-1.0.1.html)
• The annotations were downloaded as a json file and stored in the Training and Validation folders

Environment

• This code was executed in an environment which has Tensorflow 1.13.1 and Keras 2.2.5

Training Model on Custom Dataset

• Download and save the 'mask_rcnn_coco.h5' weights file in the project folder
• Refer to the custommaskrcnn.pynb notebook to see how the model was trained on 2 custom classes

Using Model Trained on Custom Dataset

• Sample outputs available in the folder