FuSeg-Net was originally created by the Milwaukee Group and M. Dhar (see original paper).
To adapt it to our use case, xfusegnet was first trained on the original data (complete FuSeg-Images, not only wound-patches) and afterwards fine-tuned for 20 epochs on manually labeled CPT-Images.
The weight-files of the original and fine-tuned xfusegnet can be found at datashare.tu-dresden.de.
For details on the training and model architecture see original_README.