Research map / Point clouds / Point cloud classification
SoPE: Spherical Coordinate-Based Positional Embedding for Enhancing Spatial Perception of 3D LVLMs
Koonting Yip, Qiyan Zhao, Wenhao Yu, Liangyu Yuen, Mingkai Li, Xiaofeng Zhang, Jianmin Ji, Yanyong Zhang, Qing Jiang, Ka-Veng Yuen
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 |
|---|---|---|---|
| replace RoPE with SoPE (spherical coordinates) in SpatialLM | SpatialLM · ARKitScenes 3D detection | F1 IoU3D@0.25 / @0.50: 63.9 -> 66.1; 60.7 -> 63.2 | Improved |
| replace RoPE with SoPE in SpatialLM | SpatialLM · Structured3D layout estimation (pretrain then finetune) | F1 IoU2D@0.25 / @0.5: 86.5 -> 88.7; 84.6 -> 86.2 | Improved |
| replace RoPE with SoPE in SpatialLM, fine-tuned on Structured3D only | SpatialLM · Structured3D layout estimation (from Structured3D only) | F1 IoU2D@0.5: 17.9 -> 29.5 | Improved |
| replace RoPE with SoPE in SpatialLM | SpatialLM · Structured3D 3D detection | F1 IoU3D@0.50: 17.6 -> 17.9 | No clear effect |
| use spherical SoPE instead of Cartesian RoPE-3D (ablation) | SpatialLM · ARKitScenes 3D detection | F1 IoU3D@0.50: 61.4 -> 63.2 | Improved |
| use MCA index reassignment projected to 2D instead of RoPE (ablation) | SpatialLM · ARKitScenes 3D detection | F1 IoU3D@0.50: 60.7 -> 60.2 | Hurt |
| use CCA index reassignment projected to 2D instead of RoPE (ablation) | SpatialLM · ARKitScenes 3D detection | F1 IoU3D@0.50: 60.7 -> 60.5 | No clear effect |
| use frequency allocation 24:2:3:3 instead of uniform 1:1:1:1 (ablation) | SpatialSoPE · ARKitScenes 3D detection | F1 IoU3D@0.50: 59.0 -> 63.2 | Improved |
| use frequency allocation 24:2:3:3 instead of temporal-biased 5:1:1:1 (ablation) | SpatialSoPE · ARKitScenes 3D detection | F1 IoU3D@0.25: 65.0 -> 66.1 | Improved |
| use frequency allocation 24:2:3:3 instead of angular-biased 8:6:9:9 (ablation) | SpatialSoPE · ARKitScenes 3D detection | F1 IoU3D@0.25: 65.5 -> 66.1 | Improved |
| add multi-scale frequency mixing to SoPE (ablation) | SpatialSoPE · ARKitScenes 3D detection | F1 IoU3D@0.50: 61.4 -> 63.2 | Improved |
| add multi-scale frequency mixing to RoPE-3D (ablation) | SpatialLM+RoPE-3D · ARKitScenes 3D detection | F1 IoU3D@0.50: 61.7 -> 62.1 | Improved |
Closest papers by meaning
- Video-3D LLM: Learning Position-Aware Video Representation for 3D Scene Understanding (CVPR 2025 · similarity 0.89)
- HoPE: Hybrid of Position Embedding for Long Context Vision-Language Models (NeurIPS 2025 · similarity 0.88)
- Uncovering and Shaping the Latent Representation of 3D Scene Topology in Vision-Language Models (NeurIPS 2026 · similarity 0.88)
- Learning from Videos for 3D World: Enhancing MLLMs with 3D Vision Geometry Priors (NeurIPS 2025 · similarity 0.88)
- S$^2$-MLLM: Boosting Spatial Reasoning Capability of MLLMs for 3D Visual Grounding with Structural Guidance (CVPR 2026 · similarity 0.87)
- From Flatland to Space: Teaching Vision-Language Models to Perceive and Reason in 3D (NeurIPS 2025 · similarity 0.87)
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.
