Image Segmentation of Metallographic Images

Microstructural segmentation using a union of attention guided U-Net models with different color transformed image

Find the original paper here.

@article{biswas2023microstructural,
  title={Microstructural segmentation using a union of attention guided U-Net models with different color transformed images},
  author={Biswas, Momojit and Pramanik, Rishav and Sen, Shibaprasad and Sinitca, Aleksandr and Kaplun, Dmitry and Sarkar, Ram},
  journal={Scientific Reports},
  volume={13},
  number={1},
  pages={5737},
  year={2023},
  publisher={Nature Publishing Group UK London}
}

Dataset Link

MetalDam

Instructions to run the code

Required directory structure: (Note: data contains subfolders of images and masks.)

+-- data
|   +-- images
|   |   +--image00
|   |   +--image01
|   |   +--image02
|   |   ...
|   +-- masks
|   |   +--mask00
|   |   +--mask01
|   |   +--mask02
|   |   ...
+-- main.py

1. Download the repository and install the required packages:

pip3 install -r requirements.txt

2. The main file is sufficient to run the experiments. Then, run the code using linux terminal as follows:

python3 main.py  --images_path "images_path" --masks_path "masks_path"

Available arguments:

• --images_path: Path where the images folder is stored. Default = ./
• --masks_path: Path where the masks folder is stored. Default = ./
• --epochs: Number of epochs of training. Default = 250
• --lr: Learning rate for training. Default = 0.001
• --batch: Batch Size for Mini Batch Training. Default = 4
• --n_splits: Number of folds for training. Default= 6
• --show: Showing the comparison among original, ground-truth and predicted images. Default = False

3. The increase_data.py is for increasing the datasize.

python3 increase_data.py --images_path "images_path" --masks_path "masks_path" --target_folder "target_folder"

Available arguments:

• --images_path: Path where the images folder is stored. Default = ./
• --masks_path: Path where the masks folder is stored. Default = ./
• --target_folder: arget folder where the images folder and the maskes folder are stored. Default = ./
• --n: Increase the data n number of times. Default = 6