Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Learning Embeddings for Image Clustering: An Empirical Study of Triplet Loss Approaches

: Ho, Kalun; Keuper, Janis; Pfreundt, Franz-Josef; Keuper, Margret


Institute of Electrical and Electronics Engineers -IEEE-; IEEE Computer Society:
ICPR 2020, 25th International Conference on Pattern Recognition. Proceedings : 10-15 January 2021, Milan, Italy, Virtual
Piscataway, NJ: IEEE, 2021
ISBN: 978-1-7281-8809-6
ISBN: 978-1-7281-8808-9
International Conference on Pattern Recognition (ICPR) <25, 2021, Online>
Deutsche Forschungsgemeinschaft DFG
KE 2264/1-1
Conference Paper
Fraunhofer ITWM ()

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 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.