We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.CV

Change to browse by:

cs

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computer Vision and Pattern Recognition

Title: Target Transformed Regression for Accurate Tracking

Abstract: Accurate tracking is still a challenging task due to appearance variations, pose and view changes, and geometric deformations of target in videos. Recent anchor-free trackers provide an efficient regression mechanism but fail to produce precise bounding box estimation. To address these issues, this paper repurposes a Transformer-alike regression branch, termed as Target Transformed Regression (TREG), for accurate anchor-free tracking. The core to our TREG is to model pair-wise relation between elements in target template and search region, and use the resulted target enhanced visual representation for accurate bounding box regression. This target contextualized representation is able to enhance the target relevant information to help precisely locate the box boundaries, and deal with the object deformation to some extent due to its local and dense matching mechanism. In addition, we devise a simple online template update mechanism to select reliable templates, increasing the robustness for appearance variations and geometric deformations of target in time. Experimental results on visual tracking benchmarks including VOT2018, VOT2019, OTB100, GOT10k, NFS, UAV123, LaSOT and TrackingNet demonstrate that TREG obtains the state-of-the-art performance, achieving a success rate of 0.640 on LaSOT, while running at around 30 FPS. The code and models will be made available at this https URL
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Cite as: arXiv:2104.00403 [cs.CV]
  (or arXiv:2104.00403v1 [cs.CV] for this version)

Submission history

From: Yutao Cui [view email]
[v1] Thu, 1 Apr 2021 11:25:23 GMT (1378kb,D)

Link back to: arXiv, form interface, contact.