search

DiamondLightSource / httomo

High-throughput tomography pipeline
https://diamondlightsource.github.io/httomo/
Other
5 stars 4 forks source link

Is `DatareducerWrapper` intended to reduce darks and flats one at a time? #268

Closed yousefmoazzam closed 6 months ago

yousefmoazzam commented 6 months ago

The current docstring: https://github.com/DiamondLightSource/httomo/blob/0bff94bf70a75c9fadbfd14cae24e62a6c66d9fd/httomo/method_wrappers/datareducer.py#L9-L12

and implementation: https://github.com/DiamondLightSource/httomo/blob/0bff94bf70a75c9fadbfd14cae24e62a6c66d9fd/httomo/method_wrappers/datareducer.py#L42-L50

suggests the title question is answered "yes"?

If so, I'm not sure if it's doing that, it looks like it may suffer a similar problem to the DezingingWrapper in #245? (In this case, keeping darks in GPU memory while flats are being transferred and processed on GPU).

yousefmoazzam commented 6 months ago

Perhaps the data reducer is only implemented as a CPU method so far, which wouldn't make this as much of an issue?

dkazanc commented 6 months ago

the current implementation will be pre-processing flats and darks on the CPU before providing the reduced arrays further to dezinger and normaliser.