Artifacts (probably from CUDA)

[sercharpak]

Good day,

I trained the network with the VERSE dataset, as indicated for 1000 epochs.
I later did inference on a separate image and got a very strange result.
I suspect it comes from CUDA, but could you confirm if you ever got this before?

Thanks!

[F4llenPotAto]

Hello Sercharpak! As a random observer, I am sorry to say I can't answer any of your questions, however, I wanted to know if any of your issues have been solved. If so, how did you do it? I'm running into the same issues as well.

[sercharpak]

Hi @F4llenPotAto , I actually re-install everything in my machine (which apparently had an issue in its motherboard), re-trained everything and it fixed this particular issue.... But I have to admit that when doing inference I still get suspicious squared segmentations in some slices... Not sure if it's related.

[F4llenPotAto]

Alright, thanks for the update. I'm currently following another lead, if that fails I'll try your method, thanks!

[sercharpak]

Have you had any chance to check this?