apple/ml-hypersim

Illumination Information

arjunsinghrathore opened this issue · 1 comments

Hi! So i wanted to know that how are you guys storing the illuminant color information for the scene? Is it for the raw hdf5 images or the tone mapped/non linear transformed image? As I was trying to save the illuminant information by saving thr input RGB values in blender. And this works for images when I'm rendering them with no non linear transformation. But when i add non linear transformation like Filmic tone mapping then the input RGB values given to blender and the illuminant color we can roughly identify from eye seems to be pretty different in some cases. So I wanted to know that how are you guys saving the illuminant if you're doing tone mapping as well?

Hi Arjun, I'm not sure I totally understand your question, but there is no tone-mapping in our HDF5 data.

Our HDF5 data adhere to the following simple equation with very low error:

color == diffuse_reflectance*diffuse_illumination + residual

If we applied a complicated tone-mapping operator to the left-hand side of this equation, it isn't obvious how to apply the same operator to the right-hand side, while also preserving our intuitive decomposition into separate images. So we intentionally avoid tone-mapping in our HDF5 data. We eventually do compute tone-mapped images because it makes the data easier to visualize, but we only save the tone-mapped images as JPG files.