Research map / Detection and segmentation / Tracking
ROAM: Recurrently Optimizing Tracking Model
Tianyu Yang, Pengfei Xu, Runbo Hu, Hua Chai, Antoni B. Chan
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 |
|---|---|---|---|
| use recurrent neural optimization instead of SGD for model updates (ablation) | ROAM · OTB-2015 | AUC: about 6% improvement | Improved |
| train the recurrent optimizer without random filter scaling (ablation) | ROAM · OTB-2015 | AUC | Hurt |
| use ground-truth boxes to construct updating samples (ablation) | ROAM and SGD · OTB-2015 | AUC | Improved |
| use the recurrent neural optimizer rather than SGD with the same oracle updating samples (ablation) | ROAM · OTB-2015 | AUC | Improved |
| replace the LSTM optimizer with a GRU optimizer (ablation) | ROAM · OTB-2015 | AUC | Hurt |
| replace the recurrent optimizer with two fully connected layers (ablation) | ROAM · OTB-2015 | AUC | Hurt |
| use a learned constant element-wise learning rate instead of an adaptive one (ablation) | ROAM · OTB-2015 | AUC | Hurt |
| use more than one gradient step at test time (ablation) | ROAM · OTB-2015 | AUC | Mixed |
| use adaptive element-wise learning rates instead of SGD with a fixed learning rate (ablation) | ROAM · OTB-2015 | AUC | Improved |
| use two gradient steps rather than one for model updates (ablation) | ROAM++ · GOT-10k test | AO | Improved |
| add bounding box regression in place of multi-scale search (ablation) | ROAM++ · GOT-10k test | AO: 0.436 -> 0.465 | Improved |
| use recurrent optimizer updates rather than no online model updating | ROAM++ · VOT-2017 | robustness: 0.195 vs 0.34 | Improved |
Closest papers by meaning
- Learning Spatio-Temporal Transformer for Visual Tracking (ICCV 2021 · similarity 0.85)
- Meta-Tracker: Fast and Robust Online Adaptation for Visual Object Trackers (ECCV 2018 · similarity 0.85)
- Alpha-Refine: Boosting Tracking Performance by Precise Bounding Box Estimation (CVPR 2021 · similarity 0.84)
- DeTrack: In-model Latent Denoising Learning for Visual Object Tracking (NeurIPS 2024 · similarity 0.84)
- MIMTrack: In-Context Tracking via Masked Image Modeling (AAAI 2025 · similarity 0.83)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.83)
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.
