Research map / Detection and segmentation / Tracking
Distractor-Aware Memory-Based Visual Object Tracking
Jovana Videnović, Matej Kristan, Alan Lukežič
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 |
|---|---|---|---|
| skip memory updates when the predicted mask is empty (ablation) | SAM2.1 · DiDi | tracking quality: 0.649 -> 0.665 | Improved |
| update memory every 5th frame instead of every frame (ablation) | SAM2.1_PRES · DiDi | tracking quality: 0.665 -> 0.667 | No clear effect |
| add DRM updated only during reliable tracking (ablation) | SAM2.1_Delta=5 · DiDi | tracking quality: 0.667 -> 0.672 | Improved |
| add DRM updated on distractor detection alone, without reliability check (ablation) | SAM2.1_Delta=5 · DiDi | tracking quality: 0.667 -> 0.644 | Hurt |
| combine distractor detection and reliability condition for DRM updates (full DAM) (ablation) | SAM2.1_Delta=5 · DiDi | tracking quality: 0.667 -> 0.694 | Improved |
| use connected components instead of a 3x3 median filter for noise removal (ablation) | DAM4SAM · DiDi | tracking quality: 0.692 -> 0.694 | No clear effect |
| remove temporal encoding from DRM (ablation) | DAM4SAM · DiDi | tracking quality: 0.669 -> 0.694 | Improved |
| always include the most recent frame in RAM (ablation) | DAM4SAM · DiDi | tracking quality: 0.685 -> 0.694 | Improved |
| use 3 RAM and 3 DRM slots instead of 4 and 4 (ablation) | DAM4SAM · VOT2022 | AO: 0.749 -> 0.792 | Improved |
| use 3 RAM and 3 DRM slots instead of 2 and 2 (ablation) | DAM4SAM · VOT2022 | AO: 0.781 -> 0.792 | Improved |
| vary theta_anc in 0.6-0.9 and theta_IoU in 0.5-0.8 (ablation) | DAM4SAM · VOT2022 | AO: Theta_anc in [0.6, 0.9]; Theta_IoU in [0.5, 0.8] | No clear effect |
| add distractor-aware memory to SAM2.1 Hiera-T | SAM2.1-T · DiDi | tracking quality: 0.600 -> 0.642 | Improved |
Closest papers by meaning
- A Distractor-Aware Memory for Visual Object Tracking with SAM2 (CVPR 2025 · similarity 0.97)
- Distractor-Aware Fast Tracking via Dynamic Convolutions and MOT Philosophy (CVPR 2021 · similarity 0.90)
- EdgeTAM: On-Device Track Anything Model (CVPR 2025 · similarity 0.87)
- When Trackers Fail: VLM-Guided Verification and Recovery for Robust Video Object Segmentation (NeurIPS 2026 · similarity 0.87)
- S2-Track: A Simple yet Strong Approach for End-to-End 3D Multi-Object Tracking (ICML 2025 · similarity 0.86)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.86)
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.
