Research map / 3D vision / 3D generation
A Tube-and-Droplet-Based Approach for Representing and Analyzing Motion Trajectories
Weiyao Lin, Yang Zhou, Hongteng Xu, Junchi Yan, Mingliang Xu, Jianxin Wu, Zicheng Liu
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 water droplet process instead of Hausdorff distance on 3D tubes (ablation) | tube-and-droplet · VMT clustering | cluster learning accuracy (%): 91.5 -> 93.8 | Improved |
| use water droplet process instead of Grassmann manifold on 3D tubes (ablation) | tube-and-droplet · VMT clustering | cluster learning accuracy (%): 93.6 -> 93.8 | Improved |
| add equipotential line extraction to build tubes instead of using raw thermal diffusion maps (ablation) | Manifold coding · VMT clustering | cluster learning accuracy (%): 92.2 -> 93.6 | Improved |
| use water droplet process instead of Hausdorff under high noise (ablation) | tube-and-droplet · VMT Noise Level 3 | cluster learning accuracy (%): 79.5 -> 88.1 | Improved |
| use water droplet process instead of Grassmann manifold under 40% trajectory breaks (ablation) | tube-and-droplet · VMT Omit G=40% | cluster learning accuracy (%): 83.2 -> 87.2 | Improved |
| use water droplet process instead of Grassmann manifold for runtime (ablation) | tube-and-droplet · VMT clustering | running time: 12.0 hr -> 47.5 sec | Improved |
| use 3D tube representation instead of pairwise or per-class trajectory models | tube-and-droplet · VMT clustering | cluster learning accuracy (%): 86.7 -> 93.8 | Improved |
| increase lambda1 (ablation) | droplet vector · trajectory A droplet (qualitative) | droplet shape | Mixed |
| use very small lambda2 (ablation) | droplet vector · trajectory A droplet (qualitative) | droplet route information | Hurt |
| use tube-and-droplet instead of tDPMM on CROSS abnormality detection | tube-and-droplet · CROSS abnormality detection | detection rate: 91.0 -> 91.3 | No clear effect |
| use tube-and-droplet instead of GPRF for classification | tube-and-droplet · CROSS classification | classification accuracy (%): 98.0 -> 98.6 | Improved |
Closest papers by meaning
- Actions as Moving Points (ECCV 2020 · similarity 0.85)
- Trace Anything: Representing Any Video in 4D via Trajectory Fields (ICLR 2026 · similarity 0.85)
- Learned Trajectory Embedding for Subspace Clustering (CVPR 2024 · similarity 0.84)
- Recovering Dynamic 3D Sketches from Videos (CVPR 2025 · similarity 0.83)
- Degrees of Freedom Matter: Inferring Dynamics from Point Trajectories (CVPR 2024 · similarity 0.82)
- Tracking and Understanding Object Transformations (NeurIPS 2025 · similarity 0.82)
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.
