Different rendering with rawpy and dcraw_emu

[pablonavarrob]

Hi everyone!

I've been playing with rawpy to write a routine to read three .RAF files, merge one given channel of each file (the red from the first, the green from the second, and the blue from the third) into a single image, and then convert them to linear .tif without having to write the single .tif files in the computer SSD, to open them again for the merging. I have managed to write a nice routine with rawpy, however, I have noticed that it produces a much different output than what I get with dcraw_emu. Is there any reason for that? Fuji files tend to be quite annoying during the demosaicing process, as sometimes "wormy" artifacts appear. Dcraw manages this really well using the AHD algorithm, however, rawpy (libraw for that matter) seems to render those artifacts regardless of the algorithm used. I haven't managed to install the algorithms "expansion pack" so I fear that might have something to do with it, however, I have no experience with C and compiling code, so I haven't been able to do it.

Below you can see the code snippet that I am using and also an image with the results.

Thanks in advance for reading this!

PARAMS_CONVERSION` = {
    'demosaic_algorithm' : rawpy.DemosaicAlgorithm(3),
    'no_auto_bright' : True,
    'output_bps' : 16,
    'gamma' : (1, 1),
    'output_color' : rawpy.ColorSpace(0),
    'user_wb' : [1, 1, 1, 1],
    'four_color_rgb' : True,
    'no_auto_bright' : True,
    'no_auto_scale' : True,
} 

i = 0
for path_red, path_green, path_blue in zip(filenames[0::3], filenames[1::3], filenames[2::3]):
    with rawpy.imread(path_red) as raw:
        red = raw.postprocess(**PARAMS_CONVERSION)

    with rawpy.imread(path_green) as raw:
        green = raw.postprocess(**PARAMS_CONVERSION)

    with rawpy.imread(path_blue) as raw:
        blue = raw.postprocess(**PARAMS_CONVERSION)

    merged_frames = np.zeros(red.shape)
    merged_frames[:, :, 0] = 0.3*red[:, :, 0]
    merged_frames[:, :, 1] = 0.59*green[:, :, 1]
    merged_frames[:, :, 2] = 0.11*blue[:, :, 2]

    i += 1
    print(i)
    imageio.imsave('frame_{}.tiff'.format(i), merged_frames.astype('uint16'))

Here's a screenshot of the different results I'm mentioning: left is demosaiced using rawpy, right is dcraw_emu, with the same algorithm (AHD). Never mind the different colors, the two images had a different ICC color profile applied in photoshop.