[Python] python OCIO colour space conversions - colour issue

I have a test 16bit tiff file that is using ACEScc colour space:

I want to process it using OCIO to, for instance, linear sRGB. OCIO provides examples of how to do it: https://opencolorio.readthedocs.io/en/latest/guides/developing/developing.html and it seems quite trivial to use:

import cv2
import numpy as np
import PyOpenColorIO as OCIO

path = "ACEScc.tif"

input_cs = "ACES - ACEScc"
output_cs = "Utility - Linear - sRGB"

img = cv2.imread(path)

#I use the default config downloaded from OCIO website
config = OCIO.Config.CreateFromFile("config.ocio")
in_colour = config.getColorSpace(input_cs)
out_colour = config.getColorSpace(output_cs)
processor = config.getProcessor(in_colour, out_colour)
cpu = processor.getDefaultCPUProcessor()

#this seems to be missing in the docs. the processor seems to expect floating point values
#scaled to 0-1 range. Here I might be doing it wrong and this might be causing the colour issues.

img2 = img.astype(np.float32) / 255.0

cpu.applyRGB(img2)

#take it back to 8bit to have something to look at:
img3 = (img2*255.0).astype(np.uint8)

cv2.imwrite(r"ocio_test.tif", img3)


the resulting image is linearised correctly. the values that are out of 8 bit range are clipped (this to be expected and I didn't handle it in any way). However the colours that are not clipped are not right. What am I missing?



here is a closeup of the macbeth chart showing the problem with the colour (left is correct, right is what I get from the above python)



some colour are ok but yellows and oranges are shifted towards pink. I also didn't expect the green plant to be out of range. I tested both with ocio v1 and v2 configs downloaded from here: https://opencolorio.readthedocs.io/en/latest/quick_start/downloads.html

I'm not sure what is the problem here and would love any suggestions.

Continue reading...