Tom Vercauteren

Yes, Gaussian derivative is more costly but it can be useful in some use cases and the best trade-off is likely application-specific.

Thanks. For python, having only a predefined functor structure with only the parameters being exposed seems quite reasonable to me and would also make it easy for user to patch...

Yes, X is a 3D vector in 3D, 2D in 2D. I fixed the missing square typo. Thanks for spotting it

Sounds great. I'm on the go at the moment but @ReubenDo in my group has been looking into our specific use case and created a "hardcoded" version for it here...

Rather than overwriting previous maual segmentation, it might be nicer to keep all the versions with a flag to inform which version should be used for a given purpose. This...

Such a feature would be great. Especially if it also works in 3D and is implemented in CUDA (or rather running efficiently on the GPU). See e.g. for a starting...

OK, I played a bit with it. For a simple 449 x 359 image and a spatial kernel diameter of 9, I ended up with these timings (the pytorch based...

Sorry, I don't have the bandwidth to do any real software development these days. If helpful, my quick and dirty notebook is here: https://colab.research.google.com/drive/1hUsuHsCD1_QOhx7yrTfSsWsylmaKRu8b?usp=sharing

This may be an upstream issue in pytorch as the following fails in all cases: ```python import torch print(f'Running PyTorch version: {torch.__version__}') torchdevices = [torch.device('cpu')] if torch.cuda.is_available(): torchdevices.append(torch.device('cuda')) print('Default GPU...

Apologies for reviving this old issue. Are there any plans to get a packaged version on pypi? This would be a nice feature to help quickly try the available wrapped...