Research map / Detection and segmentation / Tracking
Learning To Track With Object Permanence
Pavel Tokmakov, Jie Li, Wolfram Burgard, Adrien Gaidon
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 |
|---|---|---|---|
| increase ConvGRU filter size from 3x3 to 7x7 (ablation) | PermaTrack · PD val | mAP: 57.1 -> 59.6 | Improved |
| remove the H^{t-1} input (simplification) (ablation) | PermaTrack · PD val | mAP: 59.6 -> 60.6 | Improved |
| train and evaluate on videos instead of frame pairs (ablation) | PermaTrack · PD val | mAP: 60.6 -> 62.4 | Improved |
| supervise all invisible objects with unfiltered GT instead of visible-only (ablation) | PermaTrack · PD val | mAP: 62.4 -> 62.2 | Hurt |
| add annotation filtering for invisible objects (ablation) | PermaTrack · PD val | mAP: 62.2 -> 65.9 | Improved |
| use 3D constant velocity pseudo-labels instead of 2D (ablation) | PermaTrack · PD val | mAP: 65.7 -> 67.0 | Improved |
| use 3D constant velocity labels instead of filtered ground truth (ablation) | PermaTrack · PD val | mAP: 65.9 -> 67.0 | Improved |
| apply 2D constant velocity post-processing to CenterTrack (ablation) | CenterTrack · PD val | mAP: 60.3 -> 61.2 | Improved |
| apply 2D constant velocity post-processing to PermaTrack (ablation) | PermaTrack · PD val | mAP: 67.0 -> 67.9 | Improved |
| train on PD and KITTI jointly instead of fine-tuning on KITTI (ablation) | PermaTrack · KITTI val | mAP: 60.2 -> 70.5 | Improved |
| fine-tune on KITTI after PD pretraining (ablation) | PermaTrack · KITTI val Car | Car AP: 83.3 -> 75.7 | Hurt |
| fine-tune on KITTI after PD pretraining (ablation) | PermaTrack · KITTI val Person | Person AP: 38.2 -> 44.6 | Improved |
Closest papers by meaning
- Learning Object Permanence from Video (ECCV 2020 · similarity 0.90)
- Object Permanence Emerges in a Random Walk along Memory (ICML 2022 · similarity 0.90)
- Beyond Local Search: Tracking Objects Everywhere With Instance-Specific Proposals (CVPR 2016 · similarity 0.90)
- Tracking Through Containers and Occluders in the Wild (CVPR 2023 · similarity 0.90)
- DiffusionTrack: Diffusion Model for Multi-Object Tracking (AAAI 2024 · similarity 0.89)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.89)
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.
