Closed LeonBrands closed 2 years ago
It's very unlikely that the issue is in the history tracking. Please, reach me out (the email is in one header) and provide a sample app. If it's impossible, try "CommonSettings::splitScreen = 0.5" (left part of the screen will be the input). If the issue is visible in the left part of the screen in this mode - the problem is on your side, if not - the problem is in the denoiser (but still maybe input related).
This turned out to be a problem with matrix handedness that was solved in the latest version of NRD
Hi there! I'm trying to resolve some artifacts with our NRD implementation, but I'm struggling with this particular issue, and I'm hoping that someone has a clue of where this could be originated.
As seen below:
It appears that some history frames are incorrectly mapped to the screen; they are recursively added to the illumination of the center of the screen. When I pass 0 to the CommonSettings::frameIndex, this glitch doesn't appear (but ofcourse the quality is much lower without previous frame textures).
The denoiser is in the RELAX_DIFFUSE_SPECULAR mode, and my implementation roughly follows the RTXDI sample's integration, but adjusted to a DirectX12 only environment. My matrices are left-handed (+y up, +z forward, +x right) and row-major, transposed into column-major.