Reproduce “QVR3D: Query-Guided Visual Reasoning for Training-Free 3D Grounding” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jung-Hao Kao, Jing-En Huang, Jia-Wei Liao, Tzuhsuan Huang, Yu-Lun Liu, Cheng-Fu Chou, Jun-Cheng Chen

QVR3D grounds language-described objects in 3D scenes by selecting query-relevant views of candidate targets and anchors. QGMC selects frames covering their voxels, and the model uses visual and relational evidence to choose objects. In ScanRefer, Qwen3-VL-8B frame-only without GPE processes 10.16 frames and reaches 56.1% [email protected] versus scene-level MC’s 21.52 frames and 42.6%. This finding applies to that benchmark setup.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.