There is a paper, which explains how you can avoid having holes in your segmentation mask. Is there an implementation in TensorFlow or in Keras ? https://arxiv.org/pdf/1811.11718.pdf
...
↧