I am the developer of librawFX (https://github.com/lanthale/librawfx) and I am using with this way libraw. Generally all is working well and also the speed is good for reading raw files.
Nevertheless if I open Canon/Nikon file it takes 1-2 s to open the file (max. 3s). Opening a Sigma file takes around 3s. But if I open a fuji x-trans file it takes 15 - 20s.
Do I have to ship some extra libs to get here the same speed as with Sigma files ? Or is it a bug actually that it takes so long ?
Thank you in advance for your help.
Clemens
What demosaic method for X
What demosaic method for X-Trans do you use?
-- Alex Tutubalin @LibRaw LLC
I am using "libraw_dcraw_make
I am using "libraw_dcraw_make_mem_image". Maybe this is the issue. What should I use instat ?
dcraw_make_mem_image is the
dcraw_make_mem_image is the last step, applied after all postprocessing steps.
I suggest you to play with quality settings and/or half-size output. Also, OpenMP may help with default quality settings.
-- Alex Tutubalin @LibRaw LLC
X-Trans processing slow
Hi,
I'm having the same issues with libraw. Processing X-Trans files takes much longer than processing Bayer files.
I'm using the C-API version "0.19.5-Release".
I used the default params (I don't now which interpolation algorithm it is using). I tried digging into the source code to see whats going on, but couldn't figure out what exactly takes that much time in the libraw_dcraw_process call.
Here a few logs:
Opening file: "test.RAF"
libraw_open_file: 2.4638ms
libraw_unpack: 331.530498ms
libraw_dcraw_process: 18.465117336s
libraw_dcraw_make_mem_image: 258.297201ms
tiff.write_image: 64.792601ms (not libraw-function)
Opening file: "test.CR2"
libraw_open_file: 3.3209ms
libraw_unpack: 657.3599ms
libraw_dcraw_process: 720.410201ms
libraw_dcraw_make_mem_image: 63.3256ms
tiff.write_image: 50.7873ms (not libraw-function)
Has anyone looked into this problem already?
Thanks in advance
> Processing X-Trans files
> Processing X-Trans files takes much longer than processing Bayer files.
This is expected. If lower quality is acceptable for your task you may switch from (default) 3-pass Markesteijn to 1-pass or to linear approximation.
-- Alex Tutubalin @LibRaw LLC