Imraj Singh

I must've used the older config. I've just tried with the the new image and ir still does not recognise `nvidia-smi`, and when I run `sirf$ python -c 'import torch;...

It works! I just needed launch the devcontainer with slight updates: `"image": "ghcr.io/synerbi/sirf:latest-gpu", "runArgs": ["--gpus=all"],` This is in-line with what you wrote [here](https://github.com/SyneRBI/SIRF-SuperBuild/blob/master/docker/README.md), my apologies for not seeing this earlier....

I have managed to reproduce this problem in the ML_reconstruction.ipynb notebook in the pet SIRF-Exercises. For the acquired_data in the notebook I multiply by an arbitrary value i.e. "100000" ...

@KrisThielemans I am not sure if the precision loss due to the KL offset will have meaningful ramifications for assessing convergence etc? I did quickly look at the source and...

Hi both, I am looking at this at the moment. Would it make more sense to put it in SIRF or SIRF-contribs? (Also @KrisThielemans)

So I had a think, and we could add it into the common functionality: i.e. SIRF_torch.py file to "SIRF/src/common/". See here: #1305. This would be easier (and I think make...

Added a README with the some notes for a design discussion [here](https://github.com/Imraj-Singh/SIRF/blob/9df22d5295a216478979374fef20a9d456903ad6/src/torch/README.md).

@KrisThielemans, @casperdcl am I correct in thinking it's just CI left here, where we have a torch build and run the torch tests, then we can merge? Perhaps I can...

@KrisThielemans I noticed that it's good practice to add once_differentiable decorator so I have added that. Also it seemed like the PR was behind main, I have resynced.

The lpd image seems okay, but the GD comparison doesn't seem correct. The code for the GD comparison is: https://github.com/Imraj-Singh/SIRF/blob/7c29a5c893b2ecfcce67b61e4ff0b79695652f64/src/torch/tests/use_cases.py#L295C1-L338C108. Just as we see the image in the LPD, we...