1 paper · 1 filter
Simona Kocour, Assia Benbihi, Aikaterini Adam +1
Searching in and editing 3D scenes has become extremely intuitive with trainable scene representations that allow linking human concepts to elements in the scene. These operations…