NeRD++: Improved 3D-mirror symmetry learning from a single image
This work addresses the data and computational bottlenecks in 3D reconstruction for computer vision applications, representing an incremental improvement over the existing NeRD method.
The paper tackles the problem of inefficient 3D mirror symmetry learning from a single image by improving data and compute efficiency, resulting in enhanced data efficiency and faster inference speed as shown in experiments on synthetic and real-world datasets.
Many objects are naturally symmetric, and this symmetry can be exploited to infer unseen 3D properties from a single 2D image. Recently, NeRD is proposed for accurate 3D mirror plane estimation from a single image. Despite the unprecedented accuracy, it relies on large annotated datasets for training and suffers from slow inference. Here we aim to improve its data and compute efficiency. We do away with the computationally expensive 4D feature volumes and instead explicitly compute the feature correlation of the pixel correspondences across depth, thus creating a compact 3D volume. We also design multi-stage spherical convolutions to identify the optimal mirror plane on the hemisphere, whose inductive bias offers gains in data-efficiency. Experiments on both synthetic and real-world datasets show the benefit of our proposed changes for improved data efficiency and inference speed.