Created by Maanav Dalal, Jay Mody, Graham Power for Deltahacks 6. Materials Challenge 3rd place winner.
A microstructure analysis tool that automatically annotates 2d structures with grain size, distributions, phase masks, and phase fractions using image modeling.
git clone https://github.com/jaymody/microscan.git
pip install -r requirements.txt
cd flaskapp
python main.py
Then, you can open up your browser to the given localhost url, and upload an image (like the one under the sample directory sample/image_141.png). Here's a summary of our results (you can find the full list of results in results.pdf).
Left side is the original, right side is the predicted.
Orignal
Predicted Grain Outline
Individual Grain Boundaries
Our algorithms are able to predict phase fractions to an accuracy of
~1% for dark phases~5% for white phases~2% for the striped phases. (seeresults.pdf)
We also had an error rate of ~5% for grain sizes.
We also wanted to test our image modeling against real data, so we built a web scraper to scrape some online data:
Here were the results of the algorithm run on one of the scraped data points (more real results can be found in results_real.pdf):
