This Kaggle challenge stems from the HuBMAP project designed to fully map the human body. This specific project aims to identify kidney glomerulus at a cellular level from microscope images of tissue samples.
Beyond the very interesting application of image instance segmentation to medical imagery, this project presents two challenges: extremely high resolution images which leads to single file sizes in 100s of MB, and a very low number of training samples - 8 only. More interestingly, the two challenges are both issues and solutions: an approach to this project is to tile the original high resolution images into a host of smaller images which can then annotated and augmented to increase the size of the training data. These images are then fed through a Mask RCNN network to train an instance segmentation model to detect full or partial glomerulus samples in images.