Research map / Detection and segmentation / Tracking
Long-term Tracking in the Wild: a Benchmark
Jack Valmadre, Luca Bertinetto, Joao F. Henriques, Ran Tao, Andrea Vedaldi, Arnold W.M. Smeulders, Philip H.S. Torr, Efstratios Gavves
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 random re-detection logic with absent mode to SiamFC | SiamFC · OxUvA test | MaxGM: 0.313 -> 0.454 | Improved |
| add absent prediction via re-detection to SiamFC | SiamFC · OxUvA test | TNR: 0 -> 0.481 | Improved |
| add re-detection to SiamFC | SiamFC · OxUvA test | TPR: 0.391 -> 0.427 | Improved |
| add re-detection module to SiamFC versus post-hoc thresholding | SiamFC · OxUvA dev | TPR-TNR curve | No clear effect |
| apply post-hoc score thresholding to tracker scores | trackers with scores · OxUvA dev | TPR-TNR curve | Improved |
| target disappearance in the video | all baseline trackers · OxUvA test | TPR | Hurt |
| increase label frequency from 1Hz to 25Hz with fixed videos (ablation) | OTB-100 evaluation · OTB-100 | AUC std deviation: between 1Hz and 25Hz no significant difference | No clear effect |
| reduce label frequency to 0.5Hz (ablation) | OTB-100 evaluation · OTB-100 | AUC std deviation: meaningful degradation at 0.5Hz | Hurt |
| increase number of videos from 5 to 100 at 1Hz (ablation) | OTB-100 evaluation · OTB-100 | AUC std deviation: steady and significant reduction in variance | Improved |
| annotate more videos sparsely (1Hz) rather than fewer videos densely (25Hz) (ablation) | OTB-100 evaluation · OTB-100 | AUC std deviation: 4-5x smaller standard deviation | Improved |
| longer tracking time beyond first minute | most trackers · OxUvA test | TPR | Hurt |
| add 10% absent frames in the video | all trackers · OxUvA test | TPR: 0% to 10% absent | Hurt |
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 3 of 3 shared results
- Staple — Staple: Complementary Learners for Real-Time Tracking (CVPR 2016) · better on 3 of 3 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
- TrackingNet: A Large-Scale Dataset and Benchmark for Object Tracking in the Wild (ECCV 2018 · similarity 0.92)
- GOT-10k: A Large High-Diversity Benchmark for Generic Object Tracking in the Wild (TPAMI 2019 · similarity 0.92)
- TAO: A Large-Scale Benchmark for Tracking Any Object (ECCV 2020 · similarity 0.92)
- LaSOT: A High-quality Large-scale Single Object Tracking Benchmark (IJCV 2020 · similarity 0.91)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.89)
- OVT-B: A New Large-Scale Benchmark for Open-Vocabulary Multi-Object Tracking (NeurIPS 2024 · similarity 0.89)
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.
