Research map / Point clouds / Point cloud classification
Language Conditioned Spatial Relation Reasoning for 3D Object Grounding
Shizhe Chen, Pierre-Louis Guhur, Makarand Tapaswi, Cordelia Schmid, Ivan Laptev
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 |
|---|---|---|---|
| add language-conditioned spatial self-attention using pairwise distances and orientations | ViL3DRel · Nr3D, ground-truth object proposals | accuracy: 55.1 -> 64.4 | Improved |
| add language-conditioned spatial self-attention using pairwise distances and orientations | ViL3DRel · Sr3D, ground-truth object proposals | accuracy: 64.5 -> 72.8 | Improved |
| add color information (ablation) | Nr3D teacher baseline R1 · Nr3D, ground-truth object labels | accuracy: 53.5 -> 55.1 | Improved |
| add rotation augmentation (ablation) | Nr3D teacher baseline R2 · Nr3D, ground-truth object labels | accuracy: 55.1 -> 62.4 | Improved |
| add pairwise distance features (ablation) | ViL3DRel spatial attention · Nr3D, distance-only sentences | accuracy: 63.5 -> 72.6 | Improved |
| add pairwise orientation features (ablation) | ViL3DRel spatial attention · Nr3D, orientation-only sentences | accuracy: 61.2 -> 71.7 | Improved |
| replace multi-head spatial attention with single-head spatial attention (ablation) | ViL3DRel · Nr3D, ground-truth object labels | accuracy: 74.4 -> 67.7 | Hurt |
| replace sigsoftmax with bias-mode positional encoding (ablation) | ViL3DRel · Nr3D, ground-truth object labels | accuracy: 74.4 -> 55.4 | Hurt |
| replace sigsoftmax with contextual-mode positional encoding (ablation) | ViL3DRel · Nr3D, ground-truth object labels | accuracy: 74.4 -> 56.4 | Hurt |
| replace object-center coordinates with bottom-center coordinates (ablation) | ViL3DRel spatial feature computation · Nr3D | accuracy: 74.4 -> 74.4 | No clear effect |
| replace designed pairwise spatial features with bounding-box pairs and an MLP (ablation) | ViL3DRel spatial feature computation · Nr3D | accuracy: 74.4 -> 57.4 | Hurt |
| increase BERT encoder depth from 3 to 12 layers (ablation) | ViL3DRel text encoder · Nr3D | accuracy: 74.4 -> 52.3 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- 3D-SPS — 3D-SPS: Single-Stage 3D Visual Grounding via Referred Point Progressive Selection (CVPR 2022) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- From Objects to Anywhere: A Holistic Benchmark for Multi-level Visual Grounding in 3D Scenes (NeurIPS 2025 · similarity 0.91)
- Grounded 3D-Aware Spatial Vision-Language Modeling (CVPR 2026 · similarity 0.90)
- Towards CLIP-driven Language-free 3D Visual Grounding via 2D-3D Relational Enhancement and Consistency (CVPR 2024 · similarity 0.90)
- AugRefer: Advancing 3D Visual Grounding via Cross-Modal Augmentation and Spatial Relation-based Referring (AAAI 2025 · similarity 0.89)
- Chain of Semantics Programming in 3D Gaussian Splatting Representation for 3D Vision Grounding (CVPR 2025 · similarity 0.89)
- S$^2$-MLLM: Boosting Spatial Reasoning Capability of MLLMs for 3D Visual Grounding with Structural Guidance (CVPR 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.
