Timezone: »
Localization by image retrieval is inexpensive and scalable due to its simple mapping and matching techniques. The localization accuracy, however, depends on the quality of the underlying image features, often obtained using contrastive learning. Most contrastive learning strategies learn features that distinguish between different classes. In the context of localization, however, there is no natural definition of classes. Therefore, images are artificially separated into positive/negative classes with respect to the chosen anchor images, based on some geometric proximity measure. In this paper, we show why such divisions are problematic for learning localization features. We argue that any artificial division based on a proximity measure is undesirable due to the inherently ambiguous supervision for images near the proximity threshold. To avoid this problem, we propose a novel technique that uses soft positive/negative assignments of images for contrastive learning. Our soft assignment makes a gradual distinction between close and far images in both geometric and feature space. Experiments on four large-scale benchmark datasets demonstrate the superiority of our soft contrastive learning over the state-of-the-art method for retrieval-based visual localization.
Author Information
Janine Thoma (ETH Zurich)
Danda Pani Paudel (ETH Zürich)
Luc V Gool (Computer Vision Lab, ETH Zurich)
More from the Same Authors
-
2019 Poster: Gated CRF Loss for Weakly Supervised Semantic Image Segmentation »
Anton Obukhov · Stamatios Georgoulis · Dengxin Dai · Luc V Gool -
2021 : Spatial-Temporal Gated Transformersfor Efficient Video Processing »
Yawei Li · Babak Ehteshami Bejnordi · Bert Moons · Tijmen Blankevoort · Amirhossein Habibian · Radu Timofte · Luc V Gool -
2021 Poster: Revisiting Contrastive Methods for Unsupervised Learning of Visual Representations »
Wouter Van Gansbeke · Simon Vandenhende · Stamatios Georgoulis · Luc V Gool -
2020 Poster: GOCor: Bringing Globally Optimized Correspondence Volumes into Your Neural Network »
Prune Truong · Martin Danelljan · Luc V Gool · Radu Timofte -
2017 Poster: Soft-to-Hard Vector Quantization for End-to-End Learning Compressible Representations »
Eirikur Agustsson · Fabian Mentzer · Michael Tschannen · Lukas Cavigelli · Radu Timofte · Luca Benini · Luc V Gool -
2016 Poster: Dynamic Filter Networks »
Xu Jia · Bert De Brabandere · Tinne Tuytelaars · Luc V Gool -
2014 Poster: Quantized Kernel Learning for Feature Matching »
Danfeng Qin · Xuanli Chen · Matthieu Guillaumin · Luc V Gool -
2014 Poster: Self-Adaptable Templates for Feature Coding »
Xavier Boix · Gemma Roig · Salomon Diether · Luc V Gool -
2011 Poster: Learning Probabilistic Non-Linear Latent Variable Models for Tracking Complex Activities »
Angela Yao · Juergen Gall · Luc V Gool · Raquel Urtasun