This repository contains results from my bachelor's thesis, "Evaluation of Strategies for Automated Segmentation of Cryo-Volume-EM Data", written as the final project of my Biophysics studies at Masaryk University in Brno.
The thesis focused on automated segmentation of cryo-FIB-SEM organelle data using neural networks trained with DeepMIB. Several neural network architectures were compared, including 2D, 2.5D, and 3D approaches, with a focus on U-Net and DeepLabv3+ models. The study also evaluated how different training parameters influenced segmentation performance, including patch size, number of patches per micrograph, batch size, mask configuration, and number of training epochs.
This repository provides the best performing trained networks together with animations of segmentation results on the full dataset.
More details are available in the thesis PDF: bc_thesis-Evaluation_of_strategies_for_automated_segmentation_of_cryo_volume_EM_data_final.pdf.