Open ducha-aiki opened 1 year ago
Dear Dmytro, We tried kornia implementation a while ago, with the very basic setting of our method, note that our method is also very specific, i.e targeted to unsupervised learning so it may have different behaviour compared to supervised training. I may try to go and recheck kornia implementation once again, when we will prepare final code release. Thank you for your kornia library support.
@AliaksandrSiarohin I don't suggest that now kornia implementation would work, I think that your results would hold. I just want to improve our library, especially given that backprop through classical functions like PnP is our selling point. If that does not work, we have to fix that :)
Hey @AliaksandrSiarohin , I would also be interested in your findings, have you been able to prepare something to share? :pray:
Hi folks,
We have got a question regarding the report of kornia PnP not helping. Could you please share some example. Would be cool to have it toy-example, but original works as well. We would like to investigate and fix our implementation https://github.com/kornia/kornia/issues/2183
-- Best, Dmytro, kornia-team