Reproducibility issue

[mateuszbuda]

Reproducibility issue

[mateuszbuda]

Hi, thanks for trying my code.
If you run training from docker, then validation split is the same and you should get roughly the same (or very similar) dice score.

[whikwon]

I've tried using docker but model doesn't predict CS6667(2nd in validation set) well(dice score = 0.0).
I checked the data but it's OK.

[mateuszbuda]

Did you train your model from scratch or are you running inference only?

[whikwon]

I'm training model from scratch now.

[whikwon]

I've checked the inference only result and it was OK.