Research map / Detection and segmentation / Open-set and few-shot detection
Object Permanence Emerges in a Random Walk along Memory
Pavel Tokmakov, Allan Jabri, Jie Li, 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 |
|---|---|---|---|
| learn occluded-object localization with a random walk along evolving spatial memory instead of explicit invisible-object labels | RAM · PD validation | mAP (Track AP): 67.0 -> 72.0 | Improved |
| learn occluded-object localization with a random walk along evolving spatial memory instead of explicit invisible-object labels | RAM · KITTI validation | mAP (Track AP): 70.5 -> 75.7 | Improved |
| use local rather than global graph edges (ablation) | RAM · PD validation | mAP (Track AP): 66.2 -> 67.2 | Improved |
| reduce node-embedding pooling kernel from $5\times5$ to $3\times3$ with local edges (ablation) | RAM · PD validation | mAP (Track AP): 67.2 -> 67.9 | Improved |
| add label smoothing around the ground-truth center (ablation) | RAM · PD validation | mAP (Track AP): 67.9 -> 70.2 | Improved |
| add an overlap-avoidance penalty for occluded-object walker states (ablation) | RAM with label smoothing · PD validation | mAP (Track AP): 70.2 -> 72.0 | Improved |
| supervise RAM with exact ground-truth locations of invisible instances (ablation) | RAM · PD validation | mAP (Track AP): 72.0 -> 71.1 | Hurt |
| add heuristic box refinement (ablation) | RAM · LA-CATER-Moving | mean IoU: 55.5 -> 62.5; 40.1 -> 55.1; 34.4 -> 51.8 | Improved |
| add heuristic box refinement (ablation) | OPNet · LA-CATER-Moving | mean IoU: 69.3 -> 64.4; 40.0 -> 46.9; 30.8 -> 41.6 | Mixed |
| use a fixed-camera/static-location assumption for invisible-object prediction | OPNet Self-Sup. · LA-CATER Moving Carried | mean IoU: 27.5 -> 8.5 | Hurt |
Closest papers by meaning
- Learning Object Permanence from Video (ECCV 2020 · similarity 0.91)
- Learning To Track With Object Permanence (ICCV 2021 · similarity 0.90)
- Learning Object State Changes in Videos: An Open-World Perspective (CVPR 2024 · similarity 0.87)
- Object Concepts Emerge from Motion (NeurIPS 2025 · similarity 0.85)
- Using Diffusion Priors for Video Amodal Segmentation (CVPR 2025 · similarity 0.85)
- Imagine Before Go: Self-Supervised Generative Map for Object Goal Navigation (CVPR 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.
