GI noise at grazing angles (camera nearly parallel to a surface) #1091

New issue

Open

opened 2026-10-26 00:20:43 +00:00 by sky_line · 0 comments

sky_line commented

2026-10-26 00:20:43 +00:00

Owner

Owner, 2026-10-25 (dev 0979c17fa5, RTX 5090): ray-traced GI shows noise on surfaces viewed at extreme grazing angles (camera nearly parallel to the surface). Not very noticeable; deferred.

Likely area (to check when picked up): at grazing view angles each pixel covers a long, thin footprint on the surface, so neighbouring pixels sample very different world positions -- the GI reprojection / disocclusion depth and normal tests (plane-distance tolerance vs view angle), the spatial reuse radius and its geometric similarity weights, and the denoiser's a-trous depth/normal edge-stopping all become unreliable or over-reject there. Typical remedies: view-angle-aware plane-distance tolerance (scale by 1/cos), footprint-aware spatial kernels, and history acceptance that doesn't reject on depth-gradient alone.

Owner, 2026-10-25 (dev 0979c17fa5, RTX 5090): ray-traced GI shows noise on surfaces viewed at extreme grazing angles (camera nearly parallel to the surface). Not very noticeable; **deferred**. Likely area (to check when picked up): at grazing view angles each pixel covers a long, thin footprint on the surface, so neighbouring pixels sample very different world positions -- the GI reprojection / disocclusion depth and normal tests (plane-distance tolerance vs view angle), the spatial reuse radius and its geometric similarity weights, and the denoiser's a-trous depth/normal edge-stopping all become unreliable or over-reject there. Typical remedies: view-angle-aware plane-distance tolerance (scale by 1/cos), footprint-aware spatial kernels, and history acceptance that doesn't reject on depth-gradient alone.

sky_line added the

rendering

bug

labels

2026-10-26 00:20:43 +00:00

sky_line added this to the Slipstream project

2026-10-26 00:20:43 +00:00

sky_line self-assigned this

2026-10-26 02:58:27 +00:00