TY - GEN
T1 - Semi-supervised Object Detection via VC Learning
T2 - Computer Vision, ECCV 2022, Pt. XXXI
AU - Chen, CR
AU - Debattista, Kurt
AU - Han, Jiwan
A2 - Avidan, S
A2 - Brostow, G
A2 - Cisse, M
A2 - Farinella, GM
A2 - Hassner, T
A2 - Avidan, Shai
A2 - Brostow, Gabriel
A2 - Cissé, Moustapha
A2 - Farinella, Giovanni Maria
A2 - Hassner, Tal
N1 - Funding Information:
Acknowledgments. We thank China Scholarship Council for the funding. We also thank anonymous reviewers for their comments.
Publisher Copyright:
© 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2022
Y1 - 2022
N2 - Due to the costliness of labelled data in real-world applications, semi-supervised object detectors, underpinned by pseudo labelling, are appealing. However, handling confusing samples is nontrivial: discarding valuable confusing samples would compromise the model generalisation while using them for training would exacerbate the confirmation bias issue caused by inevitable mislabelling. To solve this problem, this paper proposes to use confusing samples proactively without label correction. Specifically, a virtual category (VC) is assigned to each confusing sample such that they can safely contribute to the model optimisation even without a concrete label. It is attributed to specifying the embedding distance between the training sample and the virtual category as the lower bound of the inter-class distance. Moreover, we also modify the localisation loss to allow high-quality boundaries for location regression. Extensive experiments demonstrate that the proposed VC learning significantly surpasses the state-of-the-art, especially with small amounts of available labels.
AB - Due to the costliness of labelled data in real-world applications, semi-supervised object detectors, underpinned by pseudo labelling, are appealing. However, handling confusing samples is nontrivial: discarding valuable confusing samples would compromise the model generalisation while using them for training would exacerbate the confirmation bias issue caused by inevitable mislabelling. To solve this problem, this paper proposes to use confusing samples proactively without label correction. Specifically, a virtual category (VC) is assigned to each confusing sample such that they can safely contribute to the model optimisation even without a concrete label. It is attributed to specifying the embedding distance between the training sample and the virtual category as the lower bound of the inter-class distance. Moreover, we also modify the localisation loss to allow high-quality boundaries for location regression. Extensive experiments demonstrate that the proposed VC learning significantly surpasses the state-of-the-art, especially with small amounts of available labels.
KW - Semi-supervised learning
KW - Object detection
UR - http://www.scopus.com/inward/record.url?scp=85142762627&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-19821-2_10
DO - 10.1007/978-3-031-19821-2_10
M3 - Conference Proceeding (Non-Journal item)
SN - 9783031198205
SN - 9783031198212
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 169
EP - 185
BT - Computer Vision – ECCV 2022
ER -