SDF-SRN: Learning Signed Distance 3D Object Reconstruction from Static Images

Dense 3D object reconstruction from a single image has recently witnessed advances, but supervising neural networks with ground-truth 3D is impractical due to the laborious process of creating paired-shape datasets. Recent efforts have turned to learning 3D reconstruction 3D supervision from RGB images with annotated 2D silhouettes, reducing the cost and effort of annotation. These techniques, , remain impractical as they still require multi-view annotations of the object instance during training. As a result, most experimental efforts to have been limited to synthetic datasets. In this paper, we address this and propose SDF-SRN, an approach that requires only a single view of at training time, offering greater utility for real-world scenarios. -SRN learns implicit 3D shape representations to handle arbitrary shape that may exist in the datasets. To this end, we derive a novel rendering formulation for learning signed distance functions (SDF) from 2D silhouettes. Our method outperforms the state of the art under single-view supervision settings on both synthetic and real-world.