- clone and install this package https://github.com/matterport/Mask_RCNN
- install the requirements in
Mask_RCNN/requirements.txt - The scripts use tensorflow 1.x
Annotate your dataset here: http://www.robots.ox.ac.uk/~vgg/software/via/via-1.0.6.html
Note: image should be in .jpg format
Place you annotated dataset in a folder with a name different than dataset (images + via_region_data.json)
Run split.py dataset_folder train_percentage validation_percentage test_percentage to split the data into test, train and validation (eg. split.py original_dataset 70 15 15, the sum of the three int values have to be 100)
You can run split.py without percentages, default values are 70 15 15
By default, the original dataset is kept intact, a folder named dataset will be created and it will contain the new processed dataset.
to perform data augmentation, run data_augmentation.py
Data augmentation is performed on ./dataset/train only
To train run:
python3 ./corrosion.py train --weights=coco --dataset=./dataset
This performs training on the pretrained model on the coco dataset
After training, you can find the latest model in the ./logs folder (eg. ./logs/damage20200329T2242/mask_rcnn_damage_0010.h5), perform testing using it
python3 ./corrosion.py test --weights="latest_model_path" --dataset=./dataset
This outputs the resulted images in ./output and displays the score for each image and the mean score
- Mask_RCNN Repo: https://github.com/matterport/Mask_RCNN
- Online annotator: http://www.robots.ox.ac.uk/~vgg/software/via/via-1.0.6.html
- The script the work is based on: https://github.com/priya-dwivedi/Deep-Learning/blob/master/mask_rcnn_damage_detection/custom.py "# corrosion-detection"