Learning Metrics From Teachers: Compact Networks For Image Embedding

Abstract

Metric learning networks are used to compute image embeddings, which are widely used in many applications such as image retrieval and face recognition. In this paper, we propose to use network distillation to efficiently compute image embeddings with small networks. Network distillation has been successfully applied to improve image classification, but has hardly been explored for metric learning. To do so, we propose two new loss functions that model the communication of a deep teacher network to a small student network. We evaluate our system in several datasets, including CUB-200-2011, Cars-196, Stanford Online Products and show that embeddings computed using small student networks perform significantly better than those computed using standard networks of similar size. Results on a very compact network (MobileNet-0.25), which can be used on mobile devices, show that the proposed method can greatly improve Recall@1 results from 27.5% to 44.6%. Furthermore, we investigate various asp

Stats

• citations98
• S2 citations—
• github stars76
• HF likes0
• heat score18.74
• arxiv keyyu2019learning

Related papers

• Few-shot Metric Learning: Online Adaptation Of Embedding For Retrieval (2022)8.09
• Data-efficient Ranking Distillation For Image Retrieval (2020)0.00
• Large-to-small Image Resolution Asymmetry In Deep Metric Learning (2022)9.62
• Learning Intra-batch Connections For Deep Metric Learning (2021)0.00
• Semi-supervised Deep Learning By Metric Embedding (2016)0.00
• Revisiting Metric Learning For Few-shot Image Classification (2019)14.90
• Constellation Loss: Improving The Efficiency Of Deep Metric Learning Loss Functions For Optimal Embedding (2019)0.00
• Divide And Conquer The Embedding Space For Metric Learning (2019)14.39