In plain words: From one room photo, the system picks furniture 3D models from a database and nudges their position and size until a rendering of the whole scene matches the picture, so objects correctly block each other. It rebuilds rooms automatically and improves on scene-understanding tests.
Abstract · IM2CAD
Given a single photo of a room and a large database of furniture CAD models, our goal is to reconstruct a scene that is as similar as possible to the scene depicted in the photograph, and composed of objects drawn from the database. We present a completely automatic system to address this IM2CAD problem that produces high quality results on challenging imagery from interior home design and remodeling websites. Our approach iteratively optimizes the placement and scale of objects in the room to best match scene renderings to the input photo, using image comparison metrics trained via deep convolutional neural nets. By operating jointly on the full scene at once, we account for inter-object occlusions. We also show the applicability of our method in standard scene understanding benchmarks where we obtain significant improvement.
Hamid Izadinia, Qi Shan, Steven M. Seitz
arXiv:1608.05137 · cs.CV · submitted Aug 18, 2016 · updated Apr 24, 2017
abstract · pdf · html · To appear at CVPR 2017
https://arxiv.org/pdf/1608.05137v1.pdf