DOrA: 3D Visual Grounding with Order-Aware Referring

arXiv:2403.16539v1 Announce Type: new
Abstract: 3D visual grounding aims to identify the target object within a 3D point cloud scene referred to by a natural language description. While previous works attempt to exploit the verbo-visual relation with proposed cross-modal transformers, unstructured natural utterances and scattered objects might lead to undesirable performances. In this paper, we introduce DOrA, a novel 3D visual grounding framework with Order-Aware referring. DOrA is designed to leverage Large Language Models (LLMs) to parse language description, suggesting …

abstract arxiv cloud cs.cv dora exploit identify language modal natural natural language object objects performances transformers type unstructured visual