Object-Centric Domain Randomization for 3D Shape Reconstruction in the Wild

arXiv:2403.14539v1 Announce Type: cross
Abstract: One of the biggest challenges in single-view 3D shape reconstruction in the wild is the scarcity of <3D shape, 2D image>-paired data from real-world environments. Inspired by remarkable achievements via domain randomization, we propose ObjectDR which synthesizes such paired data via a random simulation of visual variations in object appearances and backgrounds. Our data synthesis framework exploits a conditional generative model (e.g., ControlNet) to generate images conforming to spatial conditions such as 2.5D sketches, which …

2d image abstract arxiv challenges cs.ai cs.cv cs.lg data domain environments image object random randomization simulation type via view world