Research map / Detection and segmentation / Tracking
Video Annotation for Visual Tracking via Selection and Refinement
Kenan Dai, Jie Zhao, Lijun Wang, Dong Wang, Jianhua Li, Huchuan Lu, Xuesheng Qian, Xiaoyun Yang
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 forward/backward selection by T-Assess quality score (ablation) | VASR (vs forward-only) · LaSOT train-set annotation accuracy | mIoU / Acc@0.7: 0.834 -> 0.845; 90.7% -> 87.6% | Mixed |
| add failure detection to selection (ablation) | VASR selection · LaSOT train-set annotation accuracy | Acc@0.7: 87.6% -> 91.1% | Improved |
| add appearance-only refinement (ablation) | VASR · LaSOT train-set annotation accuracy | mIoU: 0.851 -> 0.845 | Hurt |
| add handcrafted geometric interpolation to visual refinement vs no refinement (ablation) | VASR · LaSOT train-set annotation accuracy | mIoU: 0.851 -> 0.853 | No clear effect |
| add learned visual-geometry refinement (VG-Refine Net) (ablation) | VASR · LaSOT train-set annotation accuracy | mIoU: 0.851 -> 0.865 | Improved |
| use LSTM instead of fully connected layers (ablation) | VASR · LaSOT train-set annotation accuracy | mIoU: 0.859 -> 0.865 | Improved |
| use LSTM instead of fully connected layers (ablation) | VASR · LaSOT train-set annotation accuracy | Err Rate: 0.34% -> 0.25% | Improved |
| use separate sequential predictors for forward and backward tracking (ablation) | T-Assess Net · LaSOT | annotation accuracy | Improved |
| use rectified accumulation aggregation instead of max/average pooling or summation (ablation) | VG-Refine Net training · LaSOT | annotation accuracy | Improved |
| train SiamFC++ on VASR instead of original TrackingNet annotations | SiamFC++ · TrackingNet test | Success: 0.754 -> 0.770 | Improved |
| train PrDiMP on VASR instead of original TrackingNet annotations | PrDiMP · GOT10K test | AO: 0.576 -> 0.589 | Improved |
| train SiamFC++ on VASR instead of original TrackingNet annotations | SiamFC++ · GOT10K test | SR0.75: 0.363 -> 0.442 | Improved |
Closest papers by meaning
- Robust Object Tracking Based on Temporal and Spatial Deep Networks (ICCV 2017 · similarity 0.87)
- Alpha-Refine: Boosting Tracking Performance by Precise Bounding Box Estimation (CVPR 2021 · similarity 0.86)
- Learning a Visual Tracker from a Single Movie without Annotation (AAAI 2019 · similarity 0.86)
- Joint Classification and Regression for Visual Tracking with Fully Convolutional Siamese Networks (IJCV 2022 · similarity 0.86)
- VastTrack: Vast Category Visual Object Tracking (NeurIPS 2024 · similarity 0.85)
- Tracktention: Leveraging Point Tracking to Attend Videos Faster and Better (CVPR 2025 · 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.
