Rel3D: A Minimally Contrastive Benchmark for Grounding Spatial Relations in 3D
This dataset provides a crucial resource for researchers and developers working on spatial relation understanding in 3D, particularly for robotics and computer vision, by offering high-quality 3D ground truth and a method to reduce dataset bias.
The paper introduces Rel3D, a large-scale, human-annotated 3D dataset for grounding spatial relations, addressing the lack of high-quality 3D ground truth in existing datasets. It demonstrates that minimally contrastive examples can diagnose issues in current relation detection models and lead to sample-efficient training.
Understanding spatial relations (e.g., "laptop on table") in visual input is important for both humans and robots. Existing datasets are insufficient as they lack large-scale, high-quality 3D ground truth information, which is critical for learning spatial relations. In this paper, we fill this gap by constructing Rel3D: the first large-scale, human-annotated dataset for grounding spatial relations in 3D. Rel3D enables quantifying the effectiveness of 3D information in predicting spatial relations on large-scale human data. Moreover, we propose minimally contrastive data collection -- a novel crowdsourcing method for reducing dataset bias. The 3D scenes in our dataset come in minimally contrastive pairs: two scenes in a pair are almost identical, but a spatial relation holds in one and fails in the other. We empirically validate that minimally contrastive examples can diagnose issues with current relation detection models as well as lead to sample-efficient training. Code and data are available at https://github.com/princeton-vl/Rel3D.