Dynamic Sampling for Deep Metric Learning
This work addresses the challenge of efficient training in deep metric learning for applications such as image retrieval and classification, though it is incremental as it builds on existing sampling methods.
The paper tackles the problem of training deep metric learning models by proposing a dynamic sampling strategy that orders training pairs from easy to hard, allowing the network to learn general boundaries early and refine details later. This approach consistently boosts performance when integrated with popular loss functions across tasks like fashion search, fine-grained classification, and person re-identification.
Deep metric learning maps visually similar images onto nearby locations and visually dissimilar images apart from each other in an embedding manifold. The learning process is mainly based on the supplied image negative and positive training pairs. In this paper, a dynamic sampling strategy is proposed to organize the training pairs in an easy-to-hard order to feed into the network. It allows the network to learn general boundaries between categories from the easy training pairs at its early stages and finalize the details of the model mainly relying on the hard training samples in the later. Compared to the existing training sample mining approaches, the hard samples are mined with little harm to the learned general model. This dynamic sampling strategy is formularized as two simple terms that are compatible with various loss functions. Consistent performance boost is observed when it is integrated with several popular loss functions on fashion search, fine-grained classification, and person re-identification tasks.