CHORUS: Learning Canonicalized 3D Human-Object Spatial Relations from Unbounded Synthesized Images | IEEE Conference Publication | IEEE Xplore