ABO: Dataset and Benchmarks for Real-World 3D Object Understanding

One way around the challenging problem of getting We introduce Amazon-Berkeley Objects (ABO), a new 3D annotations for real images is to focus only on synthetic, large-scale dataset of product images and 3D models corresponding computer-aided design (CAD) models [5, 27, 67]. This has to real household objects. We use this realistic, the advantage that the data is large in scale (as there are object-centric 3D dataset to measure the domain gap many 3D CAD models available for download online) but for single-view 3D reconstruction networks trained on synthetic objects. We also use multi-view images from ABO to most objects are untextured and there is no guarantee that measure the robustness of state-of-the-art metric learning the object may exist in the real world.