Significance of Softmax-Based Features in Comparison to Distance Metric Learning-Based Features

By Shota Horiguchi, Daiki Ikami, Kiyoharu Aizawa. 2017

70 citations recorded by the source corpus.

End-to-end distance metric learning (DML) has been applied to obtain features useful in many computer vision tasks. However, these DML studies have not provided equitable comparisons between features extracted from DML-based networks and softmax-based networks. In this paper, we present objective comparisons between these two approaches under the same network architecture.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Unsupervised Embedding Learning via Invariant and Spreading Instance Feature (2019, background): “In comparison, softmax embedding achieves competitive performance without sampling requirement [18].”
Semantic Drift Compensation for Class-Incremental Learning (2020, methodology): “We observe that the softmax leads to better classification performance for the first task, as mentioned in [13, 14].”
Metric Learning for Novelty and Anomaly Detection (2018, background): “This has already been observed in [11], where features learned by classification networks with typical softmax layers are compared with metric learning based features, with regard to several benchmark datasets.”
Divide and not forget: Ensemble of selectively trained experts in Continual Learning (2024, methodology): “…with feature regularization based on knowledge distillation (Li & Hoiem, 2016) weighted with α : L = (1 − α ) L CE + αL KD , where While we use CE for its simplicity and effective clustering (Horiguchi et al., 2019), it can be replaced with other training objectives, such as self-supervision.”