For converting the whole volume to
uint16, a histogram is needed to get min/max/percentiles.
This PR adds a utility for computing histogram chunk by chunk. The final histogram is obtained by "merging" together the partial histograms. The result is slightly less accurate than computing the histogram directly on the whole volume, but the error is negligible if things are done properly.
Integrate in configuration
Pipeline: integrate histogram process
Test on "bamboo" and "big"
Things to improve
- This feature is available only for HDF5 output (#170 (closed))
process_nameis hardcoded in most places (
local_reconstruction.py). This will be problematic when the process name will be associated with an "ID" (#169 (closed))
- histogram is computed with
numpy.histogram, which is relatively slow. For now it does the job, but we might consider implementing a faster version if needed (#172 (closed))