Mask R-CNN for damage detection on Windmill Blade

Requirements

Python 3.4, TensorFlow 1.3, Keras 2.0.8 and other common packages listed in requirements.txt.

Installation

1. Clone this repository

$ git clone http://141.252.12.43/adions025/maskrcnn.git

2. Install dependencies

  pip install -r requirements.txt

3. Run setup from the repository root directory

   python setup.py install

How to train in this project - Training Process

• you can find the class in src/main/damageDetection.py, allows to training your own dataset

$ python damageDetection.py train --dataset=/home/student_5/workspace/Mask_RCNN/dataset/ --weights=imagenet --logs=/home/student_5/workspace/Mask_RCNN/logs/

if your annotations are made by labelimg you need to use this:

convert your annotations xml to json for diferences regions:

• you can also find this file inside of dataset folder, just run this file ConverterXMLtoJson.py [moreinfo]

  $ python converterXMLtoJSON.py
  

• before run put your images .jpg and your .xml file inside /train and /val

• you need to have this structure :

• /Mask_rcnn

  • /dataset
    • /train
    • /val
    • converterXMLtoJson.py

Versioning

• You can use binary segmentation version or multiclass, just use version.sh file
• Make sure that this file has the necessary execution permissions.

  chmod +x versions.sh
  

• You will get two folders with the different code versions in the previous path.

Additional information:

• Mask R-CNN needs eggs, run you setup.py file to generate again.
• If you have installation problems, you can use the same enviroment (enviroment.yml) conda.

  $ conda env create -f environment.yml
  

• You can now activate the enviroment

  $ cconda activate myenv
  

• You can find more info about how to manage conda enviroments Creating an enviroment from an enviroment.yml file [moreinfo]

Authors

• Adonis González Godoy - object instance segmentation - NHL Stenden University