No, you really do need to just linearize it :) Even if you are linearizing with the generic default gamma, it should go a long way.

Interesting point.

The same problem with living in the sRGB space also applies to Gaussian filtering (or really any linear operation). Before you simply blur your image, or resample/resize etc., it is theoretically necessary to first linearize the image.

Find our work on inverse rendering and illumination-aware computational photography in our group web page: yaksoy.github.io/themes/

This work builds on Chris' work on intrinsic decomposition: yaksoy.github.io/ColorfulShad...

You can read the paper on our project web page: yaksoy.github.io/PhysicalReli...

Users can interactively insert different light sources including spot lights, point lights, projectors, or environmental illumination with full 3D control.

Read our press release: www.sfu.ca/fas/computin...

In this work led by Chris Careaga, we combine physically-based rendering, differentiable rendering, and neural rendering to achieve realistic relighting in any photograph in-the-wild.

Watch Chris' fancy video explainer: youtu.be/XFJCT3D8t0M