Awesome

E-FCN

This is the avaiable code for the paper "Evidential fully convolutional network for semantic segmentation" (arXiv:2103.13544).

Codes for Dempster-Shafer layer, pignistic transformation layer and utility layer are in the file "libs".

The file "E-Unet.ipynb" provides a demo about how to build, train, and interfere precise and imprecise segmantation with evidential FCN models. The file "Metrics.ipynb" provides a demo about how to compute PU, UIoU and ECE with a ready-trained evidential FCN model.

The file "weights_zoo" includes the parameters of two trained evidential FCN models that are used in the demo.

The required libraries and their version:

python == 3.7.10

tensorflow == 2.4.1.