Learning Embeddings for Image Clustering: An Empirical Study of Triplet Loss Approaches
- In this work, we evaluate two different image clustering objectives, k-means clustering and correlation clustering, in the context of Triplet Loss induced feature space embeddings. Specifically, we train a convolutional neural network to learn discriminative features by optimizing two popular versions of the Triplet Loss in order to study their clustering properties under the assumption of noisyIn this work, we evaluate two different image clustering objectives, k-means clustering and correlation clustering, in the context of Triplet Loss induced feature space embeddings. Specifically, we train a convolutional neural network to learn discriminative features by optimizing two popular versions of the Triplet Loss in order to study their clustering properties under the assumption of noisy labels. Additionally, we propose a new, simple Triplet Loss formulation, which shows desirable properties with respect to formal clustering objectives and outperforms the existing methods. We evaluate all three Triplet loss formulations for K-means and correlation clustering on the CIFAR-10 image classification dataset.…
Document Type: | Conference Proceeding |
---|---|
Conference Type: | Konferenzartikel |
Zitierlink: | https://opus.hs-offenburg.de/4633 | Bibliografische Angaben |
Title (English): | Learning Embeddings for Image Clustering: An Empirical Study of Triplet Loss Approaches |
Conference: | International Conference on Pattern Recognition : ICPR (25. : Jan. 10 2021 to Jan. 15 2021 : Milan, Italy) |
Author: | Kalun Ho, Janis KeuperStaff MemberORCiDGND, Franz-Josef Pfreundt, Margret Keuper |
Date of Publication (online): | 2020/07/06 |
Page Number: | 8 |
Language: | English | Inhaltliche Informationen |
Institutes: | Fakultät Elektrotechnik, Medizintechnik und Informatik (EMI) (ab 04/2019) |
Forschung / IMLA - Institute for Machine Learning and Analytics | |
Institutes: | Bibliografie | Formale Angaben |
Open Access: | Open Access |
Licence (German): | Urheberrechtlich geschützt |
Comment: | Preprint; Conference: accepted at ICPR 2021 (Milano) |
ArXiv Id: | http://arxiv.org/abs/2007.03123 |