We are exploring a ways to improve machine learning training for image segmentation of 3D tomography data based on seeds and in particular oats. We have run similar algorithms before on Alvis on other types of images (medical data). At this stage we plan on running some smaller runs at Alvis in order to verify that the architectures work. This is why we wish to start with this smaller proposal. We plan on applying for a medium project in the future since it is anticipated that we will need a lot more computational resources.