Uncertainty Estimation

[SMohammadi89]

Hi,

Thank you for the excellent work.

I wonder to know thst in which part of the code I can find the Uncertainty Estimation.

[CuiRuikai]

The script for estimating uncertainty was not intended for release, but I have added a new commit that includes the script I used to create the confidence map. I'm not sure if it will work properly in this git repo, so please let me know if you encounter any problems that you are unable to solve.

[SMohammadi89]

Thank you so much for the response. You mentioned that you used "what uncertainties do we need in bayesian deep learning for computer vision?" which I think you used epistemic uncertainty to capture the predictive variance. Am I right? This approach contains some dropout layers! Have you considered these layers during training, or did you only do some forward passes to capture the variance in multiple reconstructions?

[CuiRuikai]

I am only using some forward passes to capture the variance in multiple reconstructions

[SMohammadi89]

HI again. The code is running, but there is no variance in multiple reconstructions during test time for uncertainty estimation

[CuiRuikai]

The per point variance is usually not obviou enough to be notice in visualization, but there are small difference in multiple prediction of the same input. The key parameters that controls the effect of variance is step_size, noise_scale, n_step in the config file. You can change them and see the effect.