search

MIC-DKFZ / nnUNet

Apache License 2.0
5.9k stars 1.76k forks source link

Regarding the issue of including negative samples in the training set #1924

Closed hj950815 closed 6 months ago

hj950815 commented 9 months ago 
          Hey, you can ignore these prompts. They have been removed in nnunet v2 because they are only confusing people. The way you did it is correct: just add these images to your training set with empty segmentation masks

Originally posted by @FabianIsensee in https://github.com/MIC-DKFZ/nnUNet/issues/1412#issuecomment-1531493888

hj950815 commented 9 months ago

Hi, when i change oversample_foreground_percent from 0.33 to 0.7, fewer false positives, but more false negatives.It is difficult to reduce false positives while maintaining sensitivity Is the approach proposed here a more efficient approach? Add negative images to training set with empty segmentation masks Can you help me?

hj950815 commented 9 months ago

At the same time I also modified batch_size to 30, patch_size 645664

GregorKoehler commented 6 months ago

Hi @hj950815 , sorry for leaving this issue unanswered for so long! Can you provide some more context for your question? What is your setting and what are you trying to achieve?

GregorKoehler commented 6 months ago

Closing this issue for now since it's stale for roughly a week. Please feel free to re-open the issue if you're still facing this issue! If so, please provide some more context for your setting and what you're trying to achieve!