arXiv cs.CL· Mary Lynn Martin, Yifei Zhang, Martha Palmer, Maria Leonor Pacheco·· 12 小时前AI 评分22
ReRef-3D：语言指代引导 3D 物体放置的基准测试
ReRef-3D: A Benchmark for Spatial Referring Expression-Guided 3D Object Placement
AI 导读
ReRef-3D 是一个语言引导的 3D 桌面场景单物体放置基准，包含 998 个 CLEVR 衍生场景中的 33,826 条指令，覆盖 16 种放置类型及直接、一跳、两跳指代。评测显示区域监督的 PlaceIt3D 放置准确率最高约 80%，LLaVA-3D 约 70%，nearest 与 aligned 关系最难，指令改写对性能整体影响不大。
正文 · 原文
Abstract:We introduce ReRef-3D, a benchmark for language-guided single-object placement in controlled 3D tabletop scenes. It contains 33,826 instructions across 998 CLEVR-derived scenes, spanning 16 placement types and direct, one-hop, and two-hop references. Given that an instruction defines a region of acceptable placements rather than one coordinate, predictions are evaluated by relation satisfaction and physical validity. Each instruction also includes a verified natural rephrasing. We evaluate LLaVA-3D, 3D-LLM, 3D-LLaVA, LL3DA, and PlaceIt3D, with PlaceIt3D evaluated under both point-based and region-based supervision. Region-supervised PlaceIt3D achieves the highest placement accuracy at about 80% and LLaVA-3D reaches about 70%. Performance varies by relation type, but nearest and aligned remain the most difficult. Rephrasing has little effect on performance overall.
| Comments: | 18 pages, 3 figures. Submitted to ACL Rolling Review (ARR) |
| Subjects: | Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV) |
| ACM classes: | I.2.7; I.2.10 |
| Cite as: | arXiv:2608.16011 [cs.CL] |
| (or arXiv:2608.16011v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2608.16011 arXiv-issued DOI via DataCite |
Submission history
From: Mary Martin [view email]
[v1]
Mon, 17 Aug 2026 02:03:28 UTC (512 KB)
[v2]
Mon, 19 Oct 2026 21:00:21 UTC (428 KB)
来源：arXiv cs.CL · arxiv.org