When light gets absorbed, it becomes heat. This heat is then exchanged with the environment via radiation and convection, and through the object via conduction. Unlike light, heat is independent of wavelength as it is exchanged via molecular vibrations. We can observe heat at a point by capturing its thermal radiation, which is simply light at Long-Wave Infrared (LWIR, \(8-14\mu \textrm{m} )\) that depends only on the surface temperature and emissivity. Tracing back through the heat transport, we can estimate the absorbed light at that point.
In the general case, a scene point can be illuminated by multiple light sources, and the light can be scattered multiple times before reaching the camera. This makes the problem of intrinsic image decomposition ill-posed. But all the complexities of light transport are also present in absorbed light!
We simply apply energy conservation at each pixel independently. We have generalized the results to wavelength dependent albedo and camera response in the paper. Our method can be extended to non-Lambertian scenes if we can compute total scattered light at each pixel from a single visible image.
Visible Image
Estimated Albedo
Thermal Transient
Absorbed Light
Estimated Shading
Visible Image
Estimated Albedo
Thermal Transient
Absorbed Light
Estimated Shading
Visible Image
Estimated Albedo
Thermal Transient
Absorbed Light
Estimated Shading
Visible Image
Estimated Albedo
Thermal Transient
Absorbed Light
Estimated Shading
Visible Image
Estimated Albedo
Thermal Transient
Absorbed Light
Estimated Shading
Visible Image
Estimated Albedo
Thermal Transient
Absorbed Light
Estimated Shading
@InProceedings{Ramanagopal_2024_CVPR,
author = {Ramanagopal, Mani and Narayanan, Sriram and Sankaranarayanan, Aswin C. and Narasimhan, Srinivasa G.},
title = {A Theory of Joint Light and Heat Transport for Lambertian Scenes},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2024},
pages = {11924-11933}
}