Evidencial Fully Convolutional Network

[paul-bd]

Hi!

Is your feature request related to a problem? Please describe. Dealing with uncertainty of ground truth seems crucial for medical imaging.

Describe the solution you'd like An implementation of the Evidencial Fully Convolutional Network ( https://arxiv.org/pdf/2103.13544.pdf) would be great ! Which seems to require two layers :

• One Dempster Shafer Layer
• One utility layer

It has been successfully applied in medical imaging here (using MONAI as a backend): https://arxiv.org/abs/2104.13293

Best,

[Nic-Ma]

Hi @paul-bd ,

Thanks for the feature request, could you please help contribute a PR if you are interested? @wyli @ericspod , do you think this network would be helpful for other tasks?

Thanks.

[ericspod]

If this is a few new layers we can definitely consider adding it in as that seems like a simple addition. A demo notebook in the tutorials repo would be helpful too. Has the linked papers been accepted for publication? If not we should hold off for the authors to respond to feedback if there is anything to address.

[vikashg]

any updates on this @paul-bd