Is your feature request related to a problem? Please describe.
The joint bilateral filter (JBF) is similar to the bilateral filter but calculates its intensity range kernel on a second input image -- a so-called guidance image. This allows to integrate more information into the filter process and has slightly different applications than the usual bilateral filter (because you need a second guidance image for the filtering). MONAI currently does not contain a regular or differentiable JBF.

Describe the solution you'd like
I implemented a fully differentiable and trainable version of the JBF based on an analytical filter gradient (C++/CUDA). You can find the code here. We additionally published a paper (Scientific Reports) presenting and using that trainable joint bilateral filter for CT denoising.
The structure of the code is very similar to #1840 but contains the JBF algorithm with gradients with respect to its input, its guidance image, and its filter parameters. My GitHub also contains a gradcheck script to compare the analytical gradient to a numerical gradient based on torch.autograd.gradcheck.

I am happy to create a pull request similar to #5757 and include the trainable JBF as an additional filter layer. Please let me know what you think and add any questions or comments!