Research map / Detection and segmentation / Tracking
Meta-Tracker: Fast and Robust Online Adaptation for Visual Object Trackers
Eunbyung Park, Alexander C. Berg
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 meta-learned initialization (1 iteration) to CREST | CREST · VOT2016 (CREST family) | EAO: 0.283 -> 0.317 | Improved |
| add meta-learned initialization (1 iteration) to CREST | CREST-01 · VOT2016 (CREST family) | EAO: 0.259 -> 0.317 | Improved |
| add meta-learned initialization to CREST | CREST · VOT2016 (CREST family) | Robustness (failures): 1.083 -> 0.932 | Improved |
| add meta-learned initialization (1 iteration) to pyMDNet | pyMDNet-30 · VOT2016 (MDNet family) | EAO: 0.304 -> 0.314 | Improved |
| add meta-learned initialization to pyMDNet at 1 iteration | pyMDNet-01 · VOT2016 (MDNet family) | EAO: 0.119 -> 0.314 | Improved |
| add meta-learned initialization to pyMDNet | pyMDNet-30 · VOT2016 (MDNet family) | Accuracy: 0.540 -> 0.526 | Hurt |
| reduce initial-frame iterations from 30 to 1 in pyMDNet (no meta-learning) (ablation) | pyMDNet · VOT2016 (MDNet family) | EAO: 0.304 -> 0.119 | Hurt |
| reduce initial-frame iterations from 10 to 1 in CREST (no meta-learning) (ablation) | CREST · VOT2016 (CREST family) | EAO: 0.252 -> 0.259 | No clear effect |
| add meta-learning to cut initial-frame training time in MDNet | pyMDNet-30 · VOT2016 (MDNet family) | training time (s): 3.508 -> 0.124 | Improved |
| add meta-learning to cut initial-frame training time in CREST | CREST-01 · VOT2016 (CREST family) | training time (s): 0.395 -> 0.090 | Improved |
| add meta-learned initialization to improve generalization to future frames | CREST-65 · VOT2016 (CREST family) | lookahead loss: 0.30 -> 0.05 | Improved |
| increase meta-update iterations T beyond 1 (ablation) | MetaSDNet · meta-training | tracking performance | No clear effect |
Closest papers by meaning
- Real-Time Object Tracking via Meta-Learning: Efficient Model Adaptation and One-Shot Channel Pruning (AAAI 2020 · similarity 0.92)
- Deep Meta Learning for Real-Time Target-Aware Visual Tracking (ICCV 2019 · similarity 0.92)
- High-Performance Long-Term Tracking With Meta-Updater (CVPR 2020 · similarity 0.91)
- Robust Online Tracking With Meta-Updater (TPAMI 2022 · similarity 0.91)
- Tracking by Instance Detection: A Meta-Learning Approach (CVPR 2020 · similarity 0.91)
- VastTrack: Vast Category Visual Object Tracking (NeurIPS 2024 · similarity 0.87)
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.
