KOPS - The Institutional Repository of the University of Konstanz

Improving Unsupervised Label Propagation for Pose Tracking and Video Object Segmentation

Improving Unsupervised Label Propagation for Pose Tracking and Video Object Segmentation

Cite This

Files in this item

Checksum: MD5:410fa7e68045efd7a3b6e580c8dc7314

WALDMANN, Urs, Jannik BAMBERGER, Ole JOHANNSEN, Oliver DEUSSEN, Bastian GOLDLÜCKE, 2022. Improving Unsupervised Label Propagation for Pose Tracking and Video Object Segmentation. 4th DAGM German Conference on Pattern Recognition (DAGM GCPR 2022). Konstanz, Sep 27, 2022 - Sep 30, 2022. In: ANDRES, Björn, ed. and others. Pattern Recognition : 44th DAGM German Conference, DAGM GCPR 2022, Konstanz, Germany, September 27–30, 2022, Proceedings. Cham:Springer, pp. 230-245. ISSN 0302-9743. eISSN 1611-3349. ISBN 978-3-031-16787-4. Available under: doi: 10.1007/978-3-031-16788-1_15

@inproceedings{Waldmann2022Impro-58701, title={Improving Unsupervised Label Propagation for Pose Tracking and Video Object Segmentation}, year={2022}, doi={10.1007/978-3-031-16788-1_15}, number={13485}, isbn={978-3-031-16787-4}, issn={0302-9743}, address={Cham}, publisher={Springer}, series={Lecture Notes in Computer Science}, booktitle={Pattern Recognition : 44th DAGM German Conference, DAGM GCPR 2022, Konstanz, Germany, September 27–30, 2022, Proceedings}, pages={230--245}, editor={Andres, Björn}, author={Waldmann, Urs and Bamberger, Jannik and Johannsen, Ole and Deussen, Oliver and Goldlücke, Bastian} }

Goldlücke, Bastian Waldmann, Urs Deussen, Oliver Improving Unsupervised Label Propagation for Pose Tracking and Video Object Segmentation 2022 terms-of-use Johannsen, Ole Label propagation is a challenging task in computer vision with many applications. One approach is to learn representations of visual correspondence. In this paper, we study recent works on label propagation based on correspondence, carefully evaluate the effect of various aspects of their implementation, and improve upon various details. Our pipeline assembled from these best practices outperforms the previous state of the art in terms of PCK<sub>0.1</sub> on the JHMDB dataset by 6.5%. We also propose a novel joint framework for tracking and keypoint propagation, which in contrast to the core pipeline is applicable to tracking small objects and obtains results that substantially exceed the performance of the core pipeline. Finally, for VOS, we extend the core pipeline to a fully unsupervised one by initializing the first frame with the self-attention layer from DINO. Our pipeline for VOS runs online and can handle static objects. It outperforms unsupervised frameworks with these characteristics. Waldmann, Urs Johannsen, Ole eng 2022-09-28T13:26:00Z 2022-09-28T13:26:00Z Bamberger, Jannik Deussen, Oliver Goldlücke, Bastian Bamberger, Jannik

This item appears in the following Collection(s)

Search KOPS


Browse

My Account