MELON: Reconstructing 3D objects from images with unknown poses

A technique called MELON is presented in this post, which can determine object-centric camera poses and reconstruct 3D objects without the need for approximate pose initializations, complex GAN training schemes, or pre-training on labeled data.