RGB-D-Based Categorical Object Pose and Shape Estimation: Methods, Datasets, and Evaluation
This paper addresses the problem of evaluating and improving categorical object pose and shape estimation for researchers, offering a critical review and tools to enhance fairness and generalization in the field.
This work provides an overview of methods, datasets, and evaluation protocols for 6D pose and shape estimation of objects at a per-category level, proposing new metrics and annotations that reveal existing methods generalize poorly to unconstrained orientations with a bias towards upright objects.
Recently, various methods for 6D pose and shape estimation of objects at a per-category level have been proposed. This work provides an overview of the field in terms of methods, datasets, and evaluation protocols. First, an overview of existing works and their commonalities and differences is provided. Second, we take a critical look at the predominant evaluation protocol, including metrics and datasets. Based on the findings, we propose a new set of metrics, contribute new annotations for the Redwood dataset, and evaluate state-of-the-art methods in a fair comparison. The results indicate that existing methods do not generalize well to unconstrained orientations and are actually heavily biased towards objects being upright. We provide an easy-to-use evaluation toolbox with well-defined metrics, methods, and dataset interfaces, which allows evaluation and comparison with various state-of-the-art approaches (https://github.com/roym899/pose_and_shape_evaluation).