5
$\begingroup$

Does anybody knows what operation Blender does when it saves the image to .png file?

I am interested in an exact formula, it might be also the source code. What I found in source till now is RGB->sRGB conversion code but for viewport not final rendering.

I am interesting in the subject since I wanted to render images in exr and then convert them to PNG using python script. After conversion by hand i have slight different results then Blender provides natively. I wonder where this difference come from. I need to be precise since I am going to use this data as training for neural networks so I have to be precise as possible and eliminate all possible problems beforehand.

This is the code I use for conversion in python:

def RGB2sRGB2(RGBArray):
    new_rgb = RGBArray
    higher = (new_rgb > 0.0031308)
    lower = (new_rgb <= 0.0031308)
    new_rgb[higher] = (pow(new_rgb[higher], (1.0 / 2.4))) * 1.055 - 0.055
    new_rgb[lower] = 12.92 * new_rgb[lower]
    return new_rgb

Here I post my images (native blender .png file; conversion from .exr; difference map - each blue pixel means that difference is equal 1 in range 0-255):

native blender python conversion Diffmap

$\endgroup$
3
$\begingroup$

If you select default, the transform will be the sRGB OETF as applied via OCIO. If you select any of the other view transforms such as Filmic, it's an entirely different dynamic range and transfer function.

The discrepancies you are seeing may be a byproduct of the LUT approach, but I'm not sure.

Converting a scene referred EXR to a display referred PNG isn't quite as simple as using the sRGB OETF however, and YMMV depending on your needs.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.