BendingEnergyLoss

[ajoshiusc]

Is your feature request related to a problem? Please describe.
central difference approximation is used twice in the computation of double derivatives in functions
BendingEnergyLoss

The computation of double derivatives using central difference two times could be unstable numerically. The finite difference template would be too large leading to artifacts.

Describe the solution you'd like
The correct template kernel for the second difference would be [1 -2 1] for double derivative wrt same variable.
And for mixed partials, you can use forward difference approx. for example, here
http://www.holoborodko.com/pavel/2014/11/04/computing-mixed-derivatives-by-finite-differences/

[wyli]

thanks, would be interesting to compare the memory footprint as well, would you like to contribute an implementation? I think apply_filter and seperable_filtering in MONAI could be used:

MONAI/monai/networks/layers/simplelayers.py

Line 253 in f9842c0

def apply_filter(x: torch.Tensor, kernel: torch.Tensor, **kwargs) -> torch.Tensor:

MONAI/monai/networks/layers/simplelayers.py

Line 208 in f9842c0

def separable_filtering(x: torch.Tensor, kernels: list[torch.Tensor], mode: str = "zeros") -> torch.Tensor:

[ajoshiusc]

Hi Wenqi, Yes, I am interested in contributing. Should I send pull requests, or is there another way? thanks, best, Anand

…

On Mon, Feb 6, 2023 at 1:22 AM Wenqi Li ***@***.***> wrote: thanks, would be interesting to compare the memory footprint as well, would you like to contribute an implementation? I think apply_filter and seperable_filtering in MONAI could be used: https://github.com/Project-MONAI/MONAI/blob/f9842c0dda70714c826eeb8b12bbf93d987a0584/monai/networks/layers/simplelayers.py#L253 <https://urldefense.com/v3/__https://github.com/Project-MONAI/MONAI/blob/f9842c0dda70714c826eeb8b12bbf93d987a0584/monai/networks/layers/simplelayers.py*L253__;Iw!!LIr3w8kk_Xxm!ukiCuquHz-QlcdmbxCY2McilzJsBJbJRuWrH-Eo0nGXCtmrQgDjh5PocY5_MWTUarZFFom5DisMHkb_6s8-Q86dZAQ$> https://github.com/Project-MONAI/MONAI/blob/f9842c0dda70714c826eeb8b12bbf93d987a0584/monai/networks/layers/simplelayers.py#L208 <https://urldefense.com/v3/__https://github.com/Project-MONAI/MONAI/blob/f9842c0dda70714c826eeb8b12bbf93d987a0584/monai/networks/layers/simplelayers.py*L208__;Iw!!LIr3w8kk_Xxm!ukiCuquHz-QlcdmbxCY2McilzJsBJbJRuWrH-Eo0nGXCtmrQgDjh5PocY5_MWTUarZFFom5DisMHkb_6s8-F3Eg6jA$> — Reply to this email directly, view it on GitHub <https://urldefense.com/v3/__https://github.com/Project-MONAI/MONAI/issues/5939*issuecomment-1418763387__;Iw!!LIr3w8kk_Xxm!ukiCuquHz-QlcdmbxCY2McilzJsBJbJRuWrH-Eo0nGXCtmrQgDjh5PocY5_MWTUarZFFom5DisMHkb_6s886atyBrA$>, or unsubscribe <https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/ADUILF4F7VT7DS5SZQI4WWDWWC7DVANCNFSM6AAAAAAURS4UFA__;!!LIr3w8kk_Xxm!ukiCuquHz-QlcdmbxCY2McilzJsBJbJRuWrH-Eo0nGXCtmrQgDjh5PocY5_MWTUarZFFom5DisMHkb_6s8-cLss11Q$> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[wyli]

thanks, please create a pull request, there are some steps to make the project's unit tests and coding styles consistent https://github.com/Project-MONAI/MONAI/blob/dev/CONTRIBUTING.md#preparing-pull-requests