| There has been a growing concern for both data privacy and acquisition for the purpose of training robust computer vision algorithms. Often, information is located on separate data silos and it can be difficult for a machine learning engineer to consolidate all of it in a fashion that is appropriate for model development. Additionally, some of these localized data regions may not have access to a labelled ground truth, rendering conventional model training impossible. In this paper, we propose a novel architecture that can perform image segmentation in vertical federated environments. This is the first implementation of a federated architecture that can perform vertical federated image segmentation. We utilized a distributed vertical fully convolutional network architecture that is able to train on data where the segmentation maps reside on a different federate than the original image. Our architecture is able to compress the features of an image from 49,152 down to 500, allowing for more efficient communication between the top and bottom model of our federated architecture. We trained our model on 369 images from the CamVid dataset. Our model demonstrates a robust capability for accurate road detection. |
*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.