Structure from Duplicates: Neural Inverse Graphics from a Pile of Objects--Supplementary Material -- Tianhang Cheng

Our work can be seen as an attempt in such a stride. To our knowledge, this is the first effort to conduct structure from motion from a single image. V anilla COLMAP often fails to reconstruct the duplicated instances of a single image. Therefore, vanilla COLMAP cannot find stable matching points from these extreme camera distribution. However, there are still limitations of this pose estimation module.