Materials Data Science and Informatics (IAS-9)

This research focuses on leveraging image-to-image translation Generative Adversarial Networks (GANs) for self-supervised learning of electron microscopy (EM) images. By utilizing these advanced deep learning techniques, we aim to extract rich and meaningful representations directly from raw EM data, without manual annotation. The pretrained weights obtained from these GANs serve as potent feature extractors, capturing intricate details and patterns within EM images. These representations can then be applied to various downstream tasks, including semantic and instance segmentation, denoising, resolution enhancement, and super-resolution. Such an approach accelerates feature learning and enhances the generalization and adaptability of learned representations across diverse EM imaging conditions and applications, ultimately enabling automated analysis, interpretation, and understanding of complex microscopic structures and materials. For this project, we use the CEM500K dataset (Conrad et al., 2021).

Related Publications:

• B. Kazimi, K. Ruzaeva and S. Sandfeld, "Self-Supervised Learning with Generative Adversarial Networks for Electron Microscopy," 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), Seattle, WA, USA, 2024, pp. 71-81, doi: 10.1109/CVPRW63382.2024.00012.