SA6D: Self-Adaptive Few-Shot 6D Pose Estimator for Novel and Occluded Objects

TL;DR

SA6D is a novel self-adaptive few-shot 6D pose estimation method that effectively handles novel and occluded objects without needing object-centric references, improving robustness and scalability in cluttered real-world scenes.

Contribution

The paper introduces SA6D, a self-adaptive FSPE approach that does not rely on object-centric references, enabling better generalization to novel and occluded objects in cluttered environments.

Findings

SA6D outperforms existing FSPE methods on real-world datasets.

SA6D requires fewer reference images for accurate pose estimation.

SA6D effectively handles heavy occlusions and cluttered scenes.

Abstract

To enable meaningful robotic manipulation of objects in the real-world, 6D pose estimation is one of the critical aspects. Most existing approaches have difficulties to extend predictions to scenarios where novel object instances are continuously introduced, especially with heavy occlusions. In this work, we propose a few-shot pose estimation (FSPE) approach called SA6D, which uses a self-adaptive segmentation module to identify the novel target object and construct a point cloud model of the target object using only a small number of cluttered reference images. Unlike existing methods, SA6D does not require object-centric reference images or any additional object information, making it a more generalizable and scalable solution across categories. We evaluate SA6D on real-world tabletop object datasets and demonstrate that SA6D outperforms existing FSPE methods, particularly in cluttered scenes with occlusions, while requiring fewer reference images.