Is the DLSS Super Resolution output incorrect when the velocity texture scale is [-1, 1]?

[BobAndBooBoo]

Version and Platform:

• streamline v2.4.10
• GPU: RTX4070Ti Super
• Driver: 560.94

When the velocity texture scale passed to SLWrapper::Get().TagResources_General() is [-1, 1], it seems that the image becomes blurry during camera movement if sl::Constants is set as follows:

slConstants.cameraMotionIncluded = sl::Boolean::eTrue;
slConstants.mvecScale = {1.0f, 1.0f };

When I pass the velocity texture converted from [-1, 1] to pixel space to TagResources_General() and set the following:

slConstants.cameraMotionIncluded = sl::Boolean::eTrue;
slConstants.mvecScale = {1.0f / renderSize.x, 1.0f / renderSize.y};

it works correctly.

You should be able to verify this in the StreamlineSample using the following steps, but it will crash unless the issue previously reported is resolved...

1. Modify StreamlineSample.cpp in StreamlineSample. Set slConstants.cameraMotionIncluded = sl::Boolean::eFalse and slConstants.mvecScale = { 1.0f, 1.0f }.
2. Modify mvec.hlsl in Streamline to convert from pixel space to a scale of [-1, 1].
3. Modify dlssEntry.cpp in Streamline so that mvecPixelSpace = false.

[jake-nv]

DLSS-SR is expecting mvecs in screen space as described in the DLSS Programming Guide section 3.6.1. If your mvecs are in another format (not supported directly by SL), we recommend doing the conversion in your engine and have SL pass the buffer through untouched (rather than changing SL to support your desired conversion).

[BobAndBooBoo]

Upon checking, it seems that the mvecScale settings were incorrect. After setting slConstants.mvecScale = { -1.0f, -1.0f }, the issue no longer occurs when moving the camera, so I will close this issue. Thank you for your confirmation.