Open YipengHu opened 1 year ago
Thank you, would be great if this is made available in early December when monai v1.1 is announced...
At this point, a baseline model without bending energy has been trained on the public cross-institution lower-pelvic multi-structure dataset. With sensible qualitative results and an average dice score around 0.50, the performance indicates the feasibility of the task. The training of the new version with bending energy has been started to fix the significant non-smoothing behaviour. Hopefully we could get a reasonable model by then.
Thanks @YipengHu @kate-sann5100 . If meeting any issues when preparing the bundle, feel free to ask me.
I'm working on a 2d demo in case it's useful here https://github.com/Project-MONAI/model-zoo/pull/201
just checking, any update on this one?
just checking, any update on this one?
in progress - will update this month.
The training is starting to converge and visualisation seems reliable.
I am currently working on the bundle-related scripts and configs so we would be able to upload as soon as the training is finished.
Is your feature request related to a problem? Please describe. Adding an example of trained registration network
Describe the solution you'd like With standard segmentation-driven methods, weakly supervising registration, with or without unsupervised loss
Describe alternatives you've considered Unsupervised registration
Additional context Using public data set: https://zenodo.org/record/7013610
@kate-sann5100 @wyli @mathpluscode