Research map / Video / Video segmentation and tracking
AutoTrajectory: Label-free Trajectory Extraction and Prediction from Videos using Dynamic Points
Yuexin Ma, Xinge Zhu, Xinjing Cheng, Ruigang Yang, Jiming Liu, Dinesh Manocha
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 dynamic points (Gaussian heatmaps) (ablation) | AutoTrajectory instance-point extraction · Zara1 | Ins-Precision (%): 38.3% -> 58.7% | Improved |
| add dynamic points (Gaussian heatmaps) (ablation) | AutoTrajectory instance-point extraction · Zara1 | Ins-Recall (%): 44.1% -> 63.1% | Improved |
| add optical flow in aggregation (ablation) | AutoTrajectory instance-point extraction · Zara2 | Ins-Precision (%): 43.4% -> 60.3% | Improved |
| add optical flow in aggregation (ablation) | AutoTrajectory instance-point extraction · Zara2 | Ins-Recall (%): 53.1% -> 67.9% | Improved |
| add forward extractor to the cycle (ablation) | AutoTrajectory instance-point extraction · Zara1 | Ins-Precision (%): 46.8% -> 58.7% | Improved |
| add backward extractor to the cycle (ablation) | AutoTrajectory instance-point extraction · Zara1 | Ins-Precision (%): 52.1% -> 58.7% | Improved |
| add forward-backward consistency loss (ablation) | AutoTrajectory instance-point extraction · Zara1 | Ins-Precision (%): 56.2% -> 58.7% | Improved |
| add forward-backward consistency loss (ablation) | AutoTrajectory instance-point extraction · Zara2 | Ins-Recall (%): 60.4% -> 67.9% | Improved |
| train dynamic-point model on all five scenes instead of two (ablation) | AutoTrajectory instance-point extraction · Zara1 | Ins-Precision (%): 52.3% -> 58.7% | Improved |
| add extracted trajectories from Univ and Zara2 as extra training data | S-LSTM · Zara1 semi-supervised | ADE (m): 0.347 -> 0.320 | Improved |
| add extracted trajectories from Univ and Zara2 as extra training data | LSTM · Zara1 semi-supervised | FDE (m): 1.25 -> 1.094 | Improved |
| train on extracted trajectories instead of ground truth | S-LSTM · Univ | ADE (m): 0.27 -> 0.892 | Hurt |
Closest papers by meaning
- Learning Segmentation from Point Trajectories (NeurIPS 2024 · similarity 0.86)
- SingularTrajectory: Universal Trajectory Predictor Using Diffusion Model (CVPR 2024 · similarity 0.86)
- Autoregressive Sequential Pretraining for Visual Tracking (CVPR 2025 · similarity 0.85)
- Learned Trajectory Embedding for Subspace Clustering (CVPR 2024 · similarity 0.85)
- T4P: Test-Time Training of Trajectory Prediction via Masked Autoencoder and Actor-specific Token Memory (CVPR 2024 · similarity 0.85)
- Social-Transmotion: Promptable Human Trajectory Prediction (ICLR 2024 · 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.
