search

jonbarron / robust_loss_pytorch

A pytorch port of google-research/google-research/robust_loss/
Apache License 2.0
656 stars 88 forks source link

How to apply the robust loss in Monodepth Estimation？ #13

Closed gaopeng91 closed 4 years ago

gaopeng91 commented 4 years ago

Hi， Thanks for your wonderful job，I am confused on that how to apply the robust loss on monodepth Estimation？is it used to instead photo-metirc loss？could you give me more explanation about that？ Thanks

jonbarron commented 4 years ago

Hi, unfortunately we don't have a code release for the monocular depth estimation experiments of the paper (though that code is in TF anyways so it likely isn't what you're looking for). I believe that there are Pytorch implementations of SFMLearner on Github, and using this loss should be straightforward: just delete the existing multiscale photometric loss and the smoothness term and add in AdaptiveImageLossFunction on the full-res image with: scale_lo=0.01 scale_init=0.01 and default settings for the rest and it should work (you may need to fiddle with the value of wavelet_scale_base). Appendix H of the paper goes into more detail on this.

If you do end up porting this code to a PyTorch codebase and don't get the results you're expecting, please let me know and I'll do my best to help.

gaopeng91 commented 4 years ago

Thanks for your help！I will try it