defect_detection provides a python package and a command line interface for
sementically segmenting micrographs to detect defects, using the methods
described in
Karl Pazdernik, Nicole L. LaHaye, Conor M. Artman, Yuanyuan Zhu, Microstructural classification of unirradiated LiAlO2 pellets by deep learning methods, Computational Materials Science, Volume 181, 2020, 109728, ISSN 0927-0256, https://doi.org/10.1016/j.commatsci.2020.109728.
git clone https://github.com/pnnl/defect_detection
Create environment
conda create -n ttp_env python=3.9
conda activate ttp_env
Install defect_detection
cd defect_detection
pip install -e .
If recieve cuda error:
pip uninstall torch torchvision torchaudio
pip install torch==1.10.1+cu111 torchvision==0.11.2+cu111 -f https://download.pytorch.org/whl/torch_stable.html
Download or train (see below) to create a model weight and move the model weight file (in the notebook called "segnet_lr1e-04_unirradiated_Adam_new_augmentation_EWCE_5class_model_weights.pt") to:
defect_detection/notebooks/data/Best_Models/
Move the unirradiated images to:
defect_detection/notebooks/Unirradiated_images
Open the Jupyter Notebook application and navigate to defect_detection/notebooks/TTP_run_predictions.ipynb. The notebook will guide you through loading the model, generating predictions, and visualizing the results.
If you're interested in training a segmentation model, there are some more dependencies than for simple detection. You should have access to several images and their segmentation label.
Train Script
Script to train
python bsn_paper_metrics.py combo_num save_folder model_path
combo_num: integer to select the item from the list of combination of parameters (parameter combos in file)
save_folder: location to save the outputs
model_path: location model pt files are saved
Output: model pt file
The combination of parameters can be set by editing the list of models, optimizers (opts), and loss weights (losses) within the python script. The folder of images should be set within the train script under under variable 'folder'
The training images should include a corresponding labeled image with "Labeled " added to the start of the file name
If you have access to a trained segmentation model, and you'd like to generate a segmented image and obtain performance metrics and ripley results, use Performance Metrics Script. If you'd like to calibrate the images, use Calibration Script.
For both:
python bsn_paper_metrics.py combo_num save_folder model_path
python bsn_calibrate_paper.py combo_num save_folder model_path
combo_num: integer to select the item from the list of combination of parametrs (parameter combos in file)
save_folder: location to save the outputs
model_path: location model pt files are located
The combination of parameters can be set by editing the list of models, optimizers (opts), and loss weights (losses) within the python script. The folder of images should be set within the train script under variable folder.
In both python scripts, for no_running_mean (line 17): set to True if don't want to use mean from the training data (try if test imaging conditions are different than the training data imaging conditions)
In defect_detection/train/init.py, update the dict_scaling variable at line 87 with the filename and scale value in the format filename: {'scale': [px/um]}. If you are calculating the Ripley value, update the r variable at line 645 if you want to calculate Ripley with different radius values.
Calibration Script Script to calibrate. The model pt file should already exist.
Output: metrics and images for calibrated image in folder named after the model pt file
Performance Metrics Script Script to get all performance metrics and ripley results. The model pt file should already exist.
Output: performance metrics in folder named after the model pt file
Visualize Defect Comparison Script to create plots to compare defects. The user will compare the 'defects.csv' files from the best models for irradiated and unirradiated images.
alim_um: filter by this the minimum area (in um)
imu{number}: px/um scale
Area: test image height and width
conditions: root folder name (irradiated and unirradiated) where files to process are located
output_dir: location of output plots
location_files: file path to root folders (parent folder to irradiated and unirradiated folder containing performance metrics result folders)
line 42: the parameters of the model outputs for unirradiated and irradiated models (i.e. segnet, Adam, EWCE for model, opt, loss weight)
line 75: translations of image filenames to image integers
input: root directory with output from running bsn_paper_metrics.py
output: csv files and plots for average area, density, and proportion on grain boundary plots
Visualize Proportion Script to create plots to compare pixel proportions
output_dir: location of output plots
filename: name of file with proportions
line 11: translations of image filenames to image integers
input: proportions copied from files with 'performance_metrics_short.csv' suffix from the best metrics bsn_paper_metrics.py output folders from both irradiated and unirradiated
output: proportion graphs
Visualize Ripley Results Script to create plots to compare ripley results
name_type: irradiated vs unirradiated
pred_true: 'cluster_pred_True' or 'cluster_pred_False' if ripley is on the predicted or expert-labeled image
name_plots: name of the plot
output_dir: location of output files
file_dir: location of ripley folder (input)
image_dict: translations of image filenames to image integers
input: root directory with output from running bsn_paper_metrics.py
output: plots and csv files with radius above and below the 99% thresholds
Below are the licences and copyright notices of the code we included:
- PyTorch SegNet: https://github.com/say4n/pytorch-segnet/tree/master?tab=MIT-1-ov-file
- PyTorch Multi-Class Focal Loss: https://github.com/AdeelH/pytorch-multi-class-focal-loss?tab=MIT-1-ov-file
- TopoLoss: https://github.com/HuXiaoling/TopoLoss?tab=MIT-1-ov-file
- Astropy: https://github.com/astropy/astropy?tab=BSD-3-Clause-1-ov-file#readme
- Spatstat: https://rdrr.io/cran/spatstat/man/spatstat-package.html
- PyTorch UNet SegNet: https://github.com/trypag/pytorch-unet-segnet/tree/master?tab=MIT-1-ov-file
- PyTorch-SemSeg: https://github.com/meetps/pytorch-semseg/tree/master?tab=MIT-1-ov-file
Simplified BSD
Copyright 2024 Battelle Memorial Institute
Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
-
Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
-
Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Government or any agency thereof, or Battelle Memorial Institute. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof. PACIFIC NORTHWEST NATIONAL LABORATORY operated by BATTELLE for the UNITED STATES DEPARTMENT OF ENERGY under Contract DE-AC05-76RL01830