Instead of having a fixed threshold, we can consider a more adaptive approach where we don't remove the anchor if it's representative transform is still within the camera view frustum.
A better filtering logic would be to verify if the anchor is still within view.
Originally posted by @himanshunaidu in #356