Research map / Vision-language and multimodal / CLIP and zero-shot
LocateAnything3D: Vision-Language 3D Detection with Chain-of-Sight
Yunze Man, Shihao Wang, Guowen Zhang, Johan Bjorck, Liang-Yan Gui, Jim Fan, Jan Kautz, Yu-Xiong Wang, Zhiding Yu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| serialize objects near-to-far instead of random (ablation) | LocateAnything3D · Omni3D OUT | AP3D outdoor: 17.5 -> 33.1 | Improved |
| serialize objects near-to-far instead of left-to-right (ablation) | LocateAnything3D · Omni3D OUT | AP3D outdoor: 30.6 -> 33.1 | Improved |
| add 2D box step before 3D (Chain-of-Sight) instead of direct 3D (ablation) | LocateAnything3D · Omni3D OUT | AP3D outdoor: 22.7 -> 33.1 | Improved |
| emit 2D before 3D rather than 3D-then-2D (ablation) | LocateAnything3D · Omni3D OUT | AP3D outdoor: 26.2 -> 33.1 | Improved |
| decode center-size-rotation instead of rotation-size-center (ablation) | LocateAnything3D · Omni3D OUT | AP3D outdoor: 28.8 -> 33.1 | Improved |
| decode size before rotation (CSR vs CRS) (ablation) | LocateAnything3D · Omni3D OUT | AP3D outdoor: 32.9 -> 33.1 | Improved |
| predict 2D and 3D jointly end-to-end vs external 2D detector boxes | LocateAnything3D · Omni3D novel categories (zero-shot) | AP3D kitti: 25.73 -> 25.87 | No clear effect |
| predict 2D and 3D jointly vs external Grounding-DINO 2D boxes | LocateAnything3D · Omni3D novel categories (zero-shot) | AP3D SUN-RGBD: 21.07 -> 26.33 | Improved |
| LocateAnything3D vs DetAny3D given ground-truth 2D boxes | LocateAnything3D · Omni3D (full, indoor+outdoor) | AP3D: 34.38 -> 38.90 | Improved |
| LocateAnything3D vs DetAny3D with GT 2D boxes on nuScenes | LocateAnything3D · Omni3D OUT | AP3D nuScenes: 36.7 -> 33.9 | Mixed |
Closest papers by meaning
- ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images (NeurIPS 2024 · similarity 0.90)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · similarity 0.90)
- LabelAny3D: Label Any Object 3D in the Wild (NeurIPS 2025 · similarity 0.90)
- Towards 3D Objectness Learning in an Open World (NeurIPS 2025 · similarity 0.89)
- Unifying 2D and 3D Vision-Language Understanding (ICML 2025 · similarity 0.89)
- RefineAny3D: Depth Refinement as Semantic Alignment for Monocular 3D Detection (NeurIPS 2026 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
