I am using the output of a SceneCaptureComponent2D capturing from the “SceneDepth in R” source as part of an image projector system, specifically to handle occlusion. My level includes an InstancedFoliageActor with a number of FoliageInstancedStaticMeshComponents representing rocks. These instanced static meshes are not being captured in the scene depth by the capture component by default. The only method that I’ve found to have them do so is enable “Render in Main Renderer” in the scene capture component, but this is unfortunately incompatible with NVIDIA DLSS/DLAA, and causes a crash immediately upon capturing. I haven’t been able to find any references to this crashing issue online, but I assume it’s related to the “restrictions in the engine which do not allow up-scaler plugins to work on scene captures”.. (The obvious solution here would be to just not use DLSS/DLAA, but DLAA is far and away the best deferred rendering antialiasing method for VR that I’ve found).
Is there any known way to force these instanced meshes to render in the capture component scenedepth pass when not rendering in the main pass? Alternatively, is there a way to have the scene capture happen in the main rendering pass and force it to use a different upscaler/AA solution to prevent this crash?