Materials Data Segmentation Benchmark (MDSB) and Model Garden (MSDMG)

Welcome to the Future of Materials Science!

At the forefront of innovation, our Materials Data Segmentation Project is spearheading new ways to evaluate and develop materials segmentation models. This initiative is anchored by two groundbreaking components:

  • Materials Data Segmentation Benchmark (MDSB)
  • Materials Data Segmentation Model Garden (MDSMG)

These projects are open-source and welcome contributions from the global community.

Want to contribute? Fill up the Google Form.

____________________________________________________________________________

Materials Data Segmentation Benchmark (MDSB)

MDSB is our central repository, offering an extensive collection of diverse materials datasets. From metals and alloys to polymers, complete with ground truth annotations. We support a variety of imaging modalities, including microscopy and tomography. As an open-source initiative, MDSB is a collaborative platform that fosters innovation and advancement in materials analysis techniques.

____________________________________________________________________________

Materials Data Segmentation Model Garden (MDSMG)

Complementing the MDSB, MDSMG provides a collection of segmentation models specifically trained on various materials benchmark datasets. It serves as a standardized platform for testing and comparing segmentation models. Our focus on evaluation metrics ensures consistent performance assessment across models. As an open-source project, MDSMG democratizes access to powerful segmentation tools, thereby accelerating the development of new materials analysis techniques.

Together, MDSB and MDSMG create a robust ecosystem for the continuous evolution of materials segmentation research.

____________________________________________________________________________

Our Datasets and Segmentation Models

We have meticulously curated the initial collection of images from various datasets across various characterization techniques and materials, including AFM, SEM, and XCT. Each dataset comprises a mix of hardness levels, ranging from very easy to very hard, with 20% of images in each category. This diverse collection enables comprehensive evaluation and comparison of the performance of the segmentation models on the materials science-specific tasks.

Currently, we have organized six different datasets and these datasets are stored in separate components in our OSF page.

  1. LPBF (High-Speed Camera)
  2. Fractography (SEM)
  3. Crystallites (AFM)
  4. Al-Mg-Si Stress corrosion (XCT)
  5. Al Pitting corrosion (XCT)
  6. Contact corrosion (SEM)

Furthermore, for each dataset, we provide raw images of the benchmark dataset, corresponding ground truth images, segmentation models, and evaluation scripts, all organized in separate folders. These evaluation scripts can be used to assess the segmentation models on their respective benchmark datasets.

We plan to expand our collection to include millions of images, and models offering an unprecedented resource for training and developing new segmentation models.

____________________________________________________________________________

Join Us

We invite you to be part of this exciting journey. By participating in our open-source initiatives, you’ll join a global community dedicated to advancing the field of materials science. Collaborate with us, contribute to our repositories, and help shape the future of materials segmentation research.

You can fill out this Google Form to contribute your datasets and segmentation models. Let’s work together to advance the frontiers of materials science.

Explore, innovate, and lead the way with MDSB and MDSMG. Together, we can unlock new possibilities in materials science and engineering.

____________________________________________________________________________

For more information or to start exploring our resources, visit our OSF page.

Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback
Feedback