Research map / Detection and segmentation / Tracking
Learning Optimal Parameters for Multi-target Tracking with Contextual Interactions
Shaofei Wang, Charless C. Fowlkes
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 pairwise quadratic contextual costs (LP) to linear SSP (ablation) | min-cost flow tracker · KITTI tracking cross-validation, Cyclist, DPM | MOTA (%): 52.2 -> 57.2 | Improved |
| add pairwise quadratic contextual costs (LP) to linear SSP (ablation) | min-cost flow tracker · KITTI tracking cross-validation, Car, DPM | MOTA (%): 64.9 -> 65.4 | Improved |
| add pairwise quadratic contextual costs (LP) to linear SSP (ablation) | min-cost flow tracker · KITTI tracking cross-validation, Pedestrian, DPM | MOTA (%): 49.7 -> 51.2 | Improved |
| add pairwise quadratic contextual costs (LP) to linear SSP (ablation) | min-cost flow tracker · KITTI tracking cross-validation, Cyclist, Regionlet | MOTA (%): 84.9 -> 83.2 | Hurt |
| add pairwise quadratic contextual costs (LP) to linear SSP (ablation) | min-cost flow tracker · KITTI tracking cross-validation, Car, Regionlet | MOTA (%): 80.5 -> 81.0 | No clear effect |
| add pairwise quadratic contextual costs (LP) to linear SSP (ablation) | min-cost flow tracker · MOT benchmark (cross-validation, training set) | MOTA (%): 28.7 -> 28.5 | No clear effect |
| use proposed loss instead of Hamming loss (ablation) | LP model · MOT benchmark (cross-validation, training set) | MOTA (%): 25.3 -> 28.5 | Improved |
| use 2-pass DP instead of 1-pass DP (ablation) | DP inference · KITTI tracking cross-validation, Car, Regionlet | MOTA (%): 79.0 -> 80.7 | Improved |
| use 2-pass DP instead of 1-pass DP (ablation) | DP inference · KITTI tracking cross-validation, Pedestrian, Regionlet | MOTA (%): 69.3 -> 71.0 | Improved |
| use 1-pass DP instead of LP inference (ablation) | DP inference · KITTI training sequences | speedup: up to 10x faster | Improved |
| use 2-pass DP instead of LP inference (ablation) | DP inference · KITTI training sequences | speedup: up to 3x faster | Improved |
| train on all six sequences (overfit) versus cross-validation (ablation) | LP model · PETS09 and TUD-Stadtmitte (accumulated) | MOTA (%): 51.6 -> 52.6 | Improved |
Closest papers by meaning
- TrackFlow: Multi-Object tracking with Normalizing Flows (ICCV 2023 · similarity 0.88)
- Learning of Global Objective for Network Flow in Multi-Object Tracking (CVPR 2022 · similarity 0.87)
- Multi-Target Tracking by Discrete-Continuous Energy Minimization (TPAMI 2016 · similarity 0.86)
- Multi-Timescale Collaborative Tracking (TPAMI 2016 · similarity 0.86)
- Focusing on Tracks for Online Multi-Object Tracking (CVPR 2025 · similarity 0.85)
- Is Multiple Object Tracking a Matter of Specialization? (NeurIPS 2024 · 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.
