I've been trying to use this DeepLabv3+ implementation with my dataset (~1000 annotated images of the same box, out of the same video sequence): https://github.com/srihari-humbarwadi/person_segmentation_tf2.0
But I get border artifacts like this:
Any ideas what could be causing it? Note that if I use bigger batches and train for more epochs, the borders tend to get thinner but never disappear. They also appear randomly around the image.
Any clues what could be causing them and how to solve it?