Abstract

We introduce Amazon-Berkeley Objects (ABO), a new large-scale dataset ofproduct images and 3D models corresponding to real household objects. We usethis realistic, object-centric 3D dataset to measure the domain gap forsingle-view 3D reconstruction networks trained on synthetic objects. We alsouse multi-view images from ABO to measure the robustness of state-of-the-artmetric learning approaches to different camera viewpoints. Finally, leveragingthe physically-based rendering materials in ABO, we perform single- andmulti-view material estimation for a variety of complex, real-world geometries.The full dataset is available for download athttps://amazon-berkeley-objects.s3.amazonaws.com/index.html.

 

Quick Read (beta)