About the strategy - Githubissues

Lingkai-Kong / SDE-Net

Code for paper: SDE-Net: Equipping Deep Neural network with Uncertainty Estimates

Apache License 2.0

105 stars 18 forks source link

About the strategy #10

Open Lzz-kit opened 3 years ago

Lzz-kit commented 3 years ago

Hi Lingkai !

Thanks for your fantastic paper. I want to consult some questions if possible.

(1) Which strategy does this network operate:

Strategy 1 -Input Image → DNN → Output + Uncertainty Total (Aleatoric + Epistemic)
Strategy 2 -Input Image → DNN → Output + Uncertainty Aleatoric + Uncertainty Epistemic

(2) Can this work for semantic segmentation tasks?

Lingkai-Kong commented 3 years ago

Hi,

Thanks for your interest and sorry for the late response.

(1) You can simulate the trained SDE multiple times and average the predictions. Then the entropy of the averaged prediction is the total uncertainty. (2) How to calculate the aleatoric uncertainty and epistemic uncertainty respectively is illustrated in Section 3.3 (there is a uncertainty quantification paragraph). (3) I am not in the comuputer vision community and not familar with this task. Sorry about that.

Lzz-kit commented 3 years ago

Thank you very much : )