Deep Metric Learning via Lifted Structured Feature Embedding

TLDR

Learning a distance metric between example pairs is crucial for visual recognition, and recent CNN advances have enabled discriminative training of semantic feature embeddings that cluster similar items together. The authors propose a batch‑lifting algorithm that transforms vector pairwise distances into a matrix to fully exploit training batches. They optimize a novel structured prediction objective on this lifted matrix and train on a 120k‑image Stanford Online Products dataset. Experiments on CUB‑200‑2011, CARS196, and Stanford Online Products show substantial gains over prior deep embedding methods using GoogLeNet, and the code and dataset are publicly released.

Abstract

Learning the distance metric between pairs of examples is of great importance for learning and visual recognition. With the remarkable success from the state of the art convolutional neural networks, recent works [1, 31] have shown promising results on discriminatively training the networks to learn semantic feature embeddings where similar examples are mapped close to each other and dissimilar examples are mapped farther apart. In this paper, we describe an algorithm for taking full advantage of the training batches in the neural network training by lifting the vector of pairwise distances within the batch to the matrix of pairwise distances. This step enables the algorithm to learn the state of the art feature embedding by optimizing a novel structured prediction objective on the lifted problem. Additionally, we collected Stanford Online Products dataset: 120k images of 23k classes of online products for metric learning. Our experiments on the CUB-200-2011 [37], CARS196 [19], and Stanford Online Products datasets demonstrate significant improvement over existing deep feature embedding methods on all experimented embedding sizes with the GoogLeNet [33] network. The source code and the dataset are available at: https://github.com/rksltnl/ Deep-Metric-Learning-CVPR16.