Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.
And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.
Discussion of weighting of generalized Dice loss
See original GitHub issueHi,
I would like to discuss this snippet in dice.py.
I could understand that you don’t want any weight to be inf. However, is setting them to the maximum non-infinite value a good idea?
My intuition is Dice loss is designed to emphasize the scarce foregrounds. The weights become infinity when no foreground is present, which means the associated gradients are going to teach the network how to predict backgrounds. And you assign a large weight to it? I think it is weird.
Any idea? Thanks a lot.
Issue Analytics
- State:
- Created 3 years ago
- Comments:6 (3 by maintainers)
Top Results From Across the Web
Top Related Medium Post
Top Related StackOverflow Question
Troubleshoot Live Code
Top Related Reddit Thread
Top Related Hackernoon Post
Top Related Tweet
Top Related Dev.to Post
Top Related Hashnode Post
They are useful features. I will definitely try them. Thanks for the suggestion.
maybe it’s also helpful to set
include_background=Falseor consider other components from monai such as "cropforeground’ https://github.com/Project-MONAI/MONAI/blob/master/monai/transforms/croppad/array.py#L155 and balanced sampling https://github.com/Project-MONAI/MONAI/blob/master/monai/transforms/croppad/dictionary.py#L186