Research map / Detection and segmentation / Tracking
Learning Object Permanence from Video
Aviv Shamsian, Ofri Kleinfeld, Amir Globerson, Gal Chechik
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 the second LSTM module (Where is it) instead of an MLP (ablation) | OPNet · LA-CATER (detector perception) | Overall mean IoU: 78.85 -> 81.94 | Improved |
| add the second LSTM module instead of an MLP (ablation) | OPNet · LA-CATER (detector perception) | Occluded mean IoU: 55.32 -> 78.83 | Improved |
| add OPNet two-module design over Transformer + LSTM | Transformer + LSTM · LA-CATER (detector perception) | Carried mean IoU: 28.25 -> 56.04 | Improved |
| add OPNet two-module design over Transformer + LSTM | Transformer + LSTM · LA-CATER (detector perception) | Occluded mean IoU: 80.40 -> 78.83 | Mixed |
| replace detector with ground-truth boxes and visibility bits (ablation) | OPNet (LSTM + LSTM) · LA-CATER | Overall mean IoU: 81.94 -> 85.44 | Improved |
| replace detector with ground-truth boxes and visibility bits (ablation) | OPNet (LSTM + LSTM) · LA-CATER | Carried mean IoU: 56.04 -> 76.42 | Improved |
| replace detector with ground-truth boxes (ablation) | Detector + Heuristic · LA-CATER | Occluded mean IoU: 47.03 -> 30.40 | Hurt |
| remove second module under perfect perception (ablation) | OPNet (LSTM + MLP) · LA-CATER (perfect perception) | Occluded mean IoU: 67.79 -> 19.39 | Hurt |
| train with only visible-frame supervision plus consistency loss instead of full supervision (ablation) | OPNet (LSTM + LSTM) · LA-CATER | Carried mean IoU: 56.04 -> 27.50 | Hurt |
| train with only visible-frame supervision plus consistency loss (ablation) | OPNet (LSTM + MLP) · LA-CATER | Overall mean IoU: 78.85 -> 61.18 | Hurt |
| train with only visible-frame supervision plus consistency loss (ablation) | Baseline LSTM · LA-CATER | Overall mean IoU: 67.20 -> 78.09 | Improved |
| use OPNet instead of R3D + LSTM | CATER snitch localization · CATER | Accuracy: 60.2 -> 74.8 | Improved |
Closest papers by meaning
- Learning Object State Changes in Videos: An Open-World Perspective (CVPR 2024 · similarity 0.88)
- BOP-ASK: Object-Interaction Reasoning for Vision-Language Models (CVPR 2026 · similarity 0.86)
- Object Concepts Emerge from Motion (NeurIPS 2025 · similarity 0.86)
- Using Diffusion Priors for Video Amodal Segmentation (CVPR 2025 · similarity 0.86)
- Mechanisms of Object Localization in Vision-Language Models (CVPR 2026 · similarity 0.86)
- Interpreting Object-level Foundation Models via Visual Precision Search (CVPR 2025 · 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.
