demo: Improve boundsError to take camera parameters into account #786
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
While boundsError is an approximation and the precise version would compute the projected sphere bounds taking into account perspective distortion and near plane clipping, our simpler variant can work as well and is invariant to camera rotation which is a nice property in certain cases; for example in VR you might want to compute LOD based on a "third" eye, a center camera, so that the geometry is in sync between two eyes. Our boundsError implementation was mostly serviceable but did not depend on camera field of view and had some potential issues with monotonicity for cases where the camera position was inside the sphere. We now clip the distance to znear and properly scale it so that the result is a normalized screen space error (0..1, multiply by screen height to get pixels). This ensures that if the bounds.error is 0, boundsError will return 0 even if the camera is inside, which should ensure proper cluster selection for extreme closeups.
|
Also adding preliminary normal support; this is with a fairly low normal weight and ~2px error bound so shading issues remain. Note that the LOD selection here is only performed every ~10% of the video, because this was rendered in Blender by manually keyframing camera motion and cuts visibility 😅 edit oh, Blender's camera field of view is also substantially smaller than the 60 degree vertical FOV I've computed the cuts for; so that makes the pops more pronounced, as the object would have been smaller on screen with the right settings. Let's say this video is correct within a constant scaling factor of the thresholding... 0001-0110.webm |
We previously output the normal values but not normal indices; when imported, editors tend to ignore the normal data as a result.
For now we use a fairly low weight (0.5); 1.0 is generally more appropriate for full mesh simplification at least but we'll start conservative for now.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
While boundsError is an approximation and the precise version would compute the projected sphere bounds taking into account perspective distortion and near plane clipping, our simpler variant can work as well and is invariant to camera rotation which is a nice property in certain cases; for example in VR you might want to compute LOD based on a "third" eye, a center camera, so that the geometry is in sync between two eyes.
Our boundsError implementation was mostly serviceable but did not depend on camera field of view and had some potential issues with monotonicity for cases where the camera position was inside the sphere.
We now clip the distance to znear and properly scale it so that the result is a normalized screen space error (0..1, multiply by screen height to get pixels). This ensures that if the bounds.error is 0, boundsError will return 0 even if the camera is inside, which should ensure proper cluster selection for extreme closeups.
This contribution is sponsored by Valve.