In attempts to run benchmarks on real data (20GB+) using this pipeline, instances of the paganin_filter_tomopy from httomolibgpu producing NaN values have been encountered.
In this particular case, the NaN values were not initially detected to have originated from the paganin filter method due to no runtime error occurring. The runtime error encountered in rescale_to_int (occurring two sections later than the section the paganin filter was in), and it took some time to work from the runtime error in rescale_to_int all the way back to its origin in the paganin filter in the first section.
To mitigate issues like this, it may be useful to have some kind of debug checks being performed at runtime (perhaps enabled by a --debug flag in the CLI?) that check things that are obviously wrong with data, such as:
In attempts to run benchmarks on real data (20GB+) using this pipeline, instances of the
paganin_filter_tomopyfrom httomolibgpu producingNaNvalues have been encountered.In this particular case, the
NaNvalues were not initially detected to have originated from the paganin filter method due to no runtime error occurring. The runtime error encountered inrescale_to_int(occurring two sections later than the section the paganin filter was in), and it took some time to work from the runtime error inrescale_to_intall the way back to its origin in the paganin filter in the first section.To mitigate issues like this, it may be useful to have some kind of debug checks being performed at runtime (perhaps enabled by a
--debugflag in the CLI?) that check things that are obviously wrong with data, such as:NaNvaluesNaNvalues