Research map / Detection and segmentation / Tracking
Deep Reinforcement Learning with Iterative Shift for Visual Tracking
Liangliang Ren, Xin Yuan, Jiwen Lu, Ming Yang, Jie Zhou
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 iterative shift and a two-action policy (continue, stop & update) (ablation) | Shift · OTB-2015 | distance precision: 0.822 -> 0.887 | Improved |
| add iterative shift and a two-action policy (continue, stop & update) (ablation) | Shift · OTB-2015 | overlap success AUC: 0.593 -> 0.651 | Improved |
| add restart and stop & ignore actions to the policy (ablation) | Shift + IS · OTB-2015 | distance precision: 0.887 -> 0.909 | Improved |
| add restart and stop & ignore actions to the policy (ablation) | Shift + IS · OTB-2015 | overlap success AUC: 0.651 -> 0.671 | Improved |
| add the full action set with iterative shift to the one-step shift baseline (ablation) | Shift · OTB-2015 | distance precision: 0.822 -> 0.909 | Improved |
| adding actions slows the tracker (22 fps vs 10.2 fps for full DRL-IS) (ablation) | Shift · OTB-2015 | speed (fps): 22 fps -> 10.2 fps | Hurt |
| replace MDNet with DRL-IS (iterative shift) for speed | MDNet · OTB-2015 | speed (fps): 2.1 fps -> 10.2 fps | Improved |
| replace ADNet with DRL-IS (joint motion and tracking status) | ADNet · OTB-2015 | distance precision: 0.880 -> 0.909 | Improved |
| replace C-COT with DRL-IS | C-COT · Temple-Color128 | distance precision: 0.781 -> 0.818 | Improved |
| replace MDNet with DRL-IS on deformation sequences | MDNet · OTB-2015 deformation (40 sequences) | overlap success AUC: 0.614 -> 0.631 | Improved |
| replace MDNet N with DRL-IS, both initialized from ImageNet | MDNet N · VOT-2016 | robustness rank: 5.75 -> 2.70 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- SINT — SINT++: Robust Visual Tracking via Adversarial Positive Instance Generation (CVPR 2018) · better on 2 of 2 shared results
- Staple — Staple: Complementary Learners for Real-Time Tracking (CVPR 2016) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Visual Tracking via Hierarchical Deep Reinforcement Learning (AAAI 2021 · similarity 0.90)
- Online Decision Based Visual Tracking via Reinforcement Learning (NeurIPS 2020 · similarity 0.90)
- Real-time 'Actor-Critic' Tracking (ECCV 2018 · similarity 0.90)
- Action-Decision Networks for Visual Tracking With Deep Reinforcement Learning (CVPR 2017 · similarity 0.89)
- RELO: Reinforcement Learning to Localize for Visual Object Tracking (ICML 2026 · similarity 0.88)
- MIMTrack: In-Context Tracking via Masked Image Modeling (AAAI 2025 · 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.
