The Electron Microscopy Dataset offers a comprehensive view of a 5x5x5 um section from the CA1 hippocampus region of the brain, translating to a 1065x2048x1536 volume. With a voxel resolution of approximately 5x5x5 nm, the dataset has been graciously provided as structured TIF files for training and validation. Both training and test dataset are one TIF file each consisting of 165 2D slices each. In addition the dataset also has binary ground truth images that shows proper segmentation and act as output value for training and testing. Notably, our focus was drawn to this dataset due to its meticulous annotations of mitochondria in two sub-volumes, making it an instrumental tool in our line of research.
Primary Objective: Develop a segmentation model capable of effectively working with the dataset, striving to achieve the optimal Intersection over Union (IoU) score.
Secondary Objective: Progress to semantic segmentation with the aim to track specific cell bodies throughout the dataset, adding bounding boxes for object detection.