Research map / Architectures and efficiency / Attention and transformers
Motion-aware Sparse Pipeline for Lightweight Object Tracking
Qingmao Wei, Fagui Liu, Dengke Zhang, Qingze He, Quan Tang
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 Gaussian motion prior to attention-based early pruning (ablation) | MaST-tiny · LaSOT | AUC: 60.5 -> 63.8 | Improved |
| replace attention scores with MLP predictor for pruning (ablation) | MaST-tiny · LaSOT | AUC: 60.5 -> 59.4 | Hurt |
| use motion window alone instead of attention + motion (ablation) | MaST-tiny · LaSOT | AUC: 63.8 -> 62.6 | Hurt |
| one-shot token sparsification vs dense (ablation) | MaST-tiny · LaSOT | AUC: 64.0 -> 63.8 | No clear effect |
| sparsify at layer 1 instead of later layers (ablation) | MaST-tiny · LaSOT | AUC: 63.82 vs 63.96 (layer 6) | No clear effect |
| use sparse MLP head instead of dense MLP head (ablation) | MaST-tiny · LaSOT | AUC: 64.0 -> 63.8 | No clear effect |
| use sparse MLP head instead of 3x3 conv head (ablation) | MaST-tiny · LaSOT | RPi FPS: 13.8 -> 23.2 | Improved |
| use sparse global localization heads (Loc Tokens) (ablation) | MaST-tiny · LaSOT | AUC: 59.0 -> 57.9 | Hurt |
| one-shot token pruning vs resolution reduction | MaST-tiny · LaSOT | AUC: 58.5 -> 63.8 | Improved |
| one-shot token pruning vs layer pruning | MaST-tiny · LaSOT | AUC: 59.1 -> 63.8 | Improved |
| make Gaussian window dynamic with previous box size (ablation) | MaST · GOT-10k val | AUC: 79.5 -> 79.9 | Improved |
| remove motion window (importance score only) (ablation) | MaST · GOT-10k val | AUC: 79.9 -> 76.9 | Hurt |
Closest papers by meaning
- TransCenter: Transformers With Dense Representations for Multiple-Object Tracking (TPAMI 2022 · similarity 0.87)
- UTPTrack: Towards Simple and Unified Token Pruning for Visual Tracking (CVPR 2026 · similarity 0.86)
- Less Is More: Token Context-Aware Learning for Object Tracking (AAAI 2025 · similarity 0.86)
- Sampling-Resilient Multi-Object Tracking (AAAI 2024 · similarity 0.86)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.85)
- ODTrack: Online Dense Temporal Token Learning for Visual Tracking (AAAI 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.
