/car-damage-detection-using-Mask-RCNN

Primary LanguageJupyter NotebookOtherNOASSERTION

Car Damage Detection and Segmentation using Mask R-CNN

This is a project of detecting damages from the images of car. The codes are based on implementation of Mask R-CNN by (https://github.com/matterport/Mask_RCNN) on Python 3, Keras, and TensorFlow. The model generates bounding boxes and segmentation masks for each instance of an object in the image. It's based on Feature Pyramid Network (FPN) and a ResNet101 backbone.

The repository includes:

  • Source code of Mask R-CNN built on FPN and ResNet101.
  • Instruction and Training code for Car Damage Detection
  • Jupyter notebooks to visualize the detection pipeline at every step
  • Example of training on your own dataset

Collecting Data

  • I have collected all the images from google images.
  • I split the images in 223 train images and 15 validation images.
  • I annotate the images using VIA (VGG Image Annotator) because of its simplicity.It’s a single HTML file that you download and open in a browser.
  • The VIA tool saves the annotations in a JSON file, and each mask is a set of polygon points.

Training

  • I make the file car_damage.py using the instruction from this blogpost.
  • I used Transfer Learning, which simply means that, instead of training a model from scratch, I start with a weights file that’s been trained on the COCO dataset.
  • Mask R-CNN is a fairly large model. Especially that our implementation uses ResNet101 and FPN. So you need a modern GPU with 12GB of memory. I used Google Colab's Tesla K80 GPU to train this model.
  • I train the model for 100 epochs, after which I got 0.1894 - training loss and 1.9137 - validation loss.

You can find the training steps in car_damage_training.ipynb.

Result