ABEM
Attention-based Segmentation and Reconstruction of Macromolecular Structures in Electron Microscopy Data
Within this project, attention-based deep learning techniques are developed to reduce the required amounts of annotated data in the analaysis of electron microscopy data, whereby the attention mechanism is supported by unsupervised denoising techniques and super-resolution approaches. As these techniques do not need any annotated data, they are applicable to tasks, where user annotations are nearly impossible to be acquired, e.g., detection of the 4D orientation of macromolecular structures. We will analyze our techniques by examining the outcomes when applied to images of Human betaherpesvirus 5/Human cytomegalovirus (HCMV) and Zika virus (ZIKV) virions in the cell. The PIs on this project are Timo Ropinski and Pedro Hermosilla from the Visual Computing Group, and Clarissa Read from the Central Facility for Electron Microscopy.