Research map / 3D vision / 3D generation
Communication-Efficient Multi-Vehicle Collaborative Semantic Segmentation via Sparse 3D Gaussian Sharing
Tianyu Hong, Xiaobo Zhou, Wenkai Hu, Qi Xie, Zhihui Ke, Tie Qiu
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 Semantic Gaussian Compression (learnable codebook indices) on top of selection (ablation) | GSCOOP · OPV2V | communication volume (KB) and vehicle IoU: 34.18 -> 15.87 KB; vehicle IoU 57.39 -> 57.21 | Improved |
| combine relevance-projection and semantic-uncertainty priority scores for Gaussian selection (ablation) | GSCOOP · OPV2V | vehicle IoU: 56.84 -> 57.39 (vs first single-module row), 55.20 -> 57.39 (vs second) | Improved |
| use codebook size 64 instead of 8 (ablation) | GSCOOP · OPV2V | vehicle IoU: 53.31 -> 57.21 | Improved |
| increase codebook size from 64 to 256 (ablation) | GSCOOP · OPV2V | vehicle IoU: 57.21 -> 54.79 | Hurt |
| increase codebook size from 64 to 128 (ablation) | GSCOOP · V2X-Seq | avg IoU: 60.49 -> 58.15 | Hurt |
| use 100*100*4 initial Gaussian voxels instead of 100*100*8 (ablation) | GSCOOP · OPV2V | vehicle IoU: 55.33 -> 57.21 | Improved |
| use 100*100*4 initial Gaussian voxels instead of 80*80*4 (ablation) | GSCOOP · V2X-Seq | avg IoU: 54.63 -> 60.49 | Improved |
| add scale and rotation attributes to transmitted Gaussians (position only vs all) (ablation) | GSCOOP Gaussian attributes · OPV2V | vehicle IoU and volume (KB): 57.21 -> 58.06 IoU; 15.87 -> 50.05 KB | No clear effect |
| transmit position attributes versus semantics only (ablation) | GSCOOP Gaussian attributes · OPV2V | vehicle IoU and volume (KB): 52.38 -> 57.21 IoU; 1.22 -> 15.87 KB | Improved |
| sparse Gaussian sharing instead of dense BEV feature sharing (GSCOOP vs CoBEVT) | collaborative BEV segmentation · OPV2V | avg IoU: 49.90 -> 56.20 | Improved |
| sparse Gaussian sharing instead of dense BEV feature sharing (GSCOOP vs ERMVP) | collaborative BEV segmentation · V2X-Seq | avg IoU: 56.48 -> 60.49 | Improved |
| Gaussian-based fusion instead of BEV feature fusion network (GSCOOP vs CoBEVT) | inference time with 7 vehicles · OPV2V | inference time (ms): 511.32 -> 242.35 ms | Improved |
Closest papers by meaning
- Vision-Only Gaussian Splatting for Collaborative Semantic Occupancy Prediction (AAAI 2026 · similarity 0.88)
- Generative Map Priors for Collaborative BEV Semantic Segmentation (CVPR 2025 · similarity 0.88)
- InstanceGaussian: Appearance-Semantic Joint Gaussian Representation for 3D Instance-Level Perception (CVPR 2025 · similarity 0.86)
- GaussianFormer-2: Probabilistic Gaussian Superposition for Efficient 3D Occupancy Prediction (CVPR 2025 · similarity 0.85)
- Collaborative Semantic Occupancy Prediction with Hybrid Feature Fusion in Connected Automated Vehicles (CVPR 2024 · similarity 0.85)
- AGC-Drive: A Large-Scale Dataset for Real-World Aerial-Ground Collaboration in Driving Scenarios (NeurIPS 2025 · similarity 0.85)
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.
