Research map / Learning theory and algorithms / Optimal transport
2D Gaussians Spatial Transport for Point-supervised Density Regression
Miao Shang, Xiaopeng Hong
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 |
|---|---|---|---|
| use heuristic Gaussian transport kernel instead of L2 direct regression (ablation) | VGG counting model · JHU-Crowd++ (VGG) | MAE: 81.70 -> 65.45 | Improved |
| derive transport kernel from 2D Gaussian splatting instead of heuristic (ablation) | GST · JHU-Crowd++ (VGG) | MAE: 65.45 -> 60.17 | Improved |
| add pixel-to-background correspondence (ablation) | GST · JHU-Crowd++ (VGG) | MAE: 60.17 -> 58.30 | Improved |
| replace OT loss with GST loss | VGG counting model · UCF-QNRF (VGG) | MAE: 87.2 -> 82.5 | Improved |
| replace OT loss with GST loss | Transformer counting model · NWPU-Crowd | MAE: 80.9 -> 74.4 | Improved |
| replace OT loss with GST (training time) | VGG counting model · JHU-Crowd++ (VGG) | training time: 28h36min -> 15h32min | Improved |
| use GST instead of PET/APGCC on UCF-QNRF | Transformer counting model · UCF-QNRF | MAE: 79.5 -> 80.7 | Hurt |
| use ground-truth scale information for the plan instead of 2DGS scales (ablation) | GST · JHU-Crowd++ (VGG) | MAE: 58.30 -> 56.60 | Improved |
| use GST instead of HRNet direct regression | HRNet-W32 · MPII Human Pose | mean PCKh@0.5: 90.4 -> 91.0 | Improved |
| use GST instead of HRNet on head joint | HRNet-W32 · MPII Human Pose | Head PCKh@0.5: 97.1 -> 96.2 | Hurt |
Closest papers by meaning
- Gaussian Density Splatting Network (NeurIPS 2026 · similarity 0.86)
- Gaussian Graph Network: Learning Efficient and Generalizable Gaussian Representations from Multi-view Images (NeurIPS 2024 · similarity 0.85)
- Gaussian Splatting Lucas-Kanade (ICLR 2025 · similarity 0.85)
- DeSiRe-GS: 4D Street Gaussians for Static-Dynamic Decomposition and Surface Reconstruction for Urban Driving Scenes (CVPR 2025 · similarity 0.84)
- GS-SLAM: Dense Visual SLAM with 3D Gaussian Splatting (CVPR 2024 · similarity 0.84)
- VoxelSplat: Dynamic Gaussian Splatting as an Effective Loss for Occupancy and Flow Prediction (CVPR 2025 · similarity 0.84)
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.
