PT Unknown
AU Lu Yu
   Vacit Oguz Yazici
   Xialei Liu
   Joost Van de Weijer
   Yongmei Cheng
   Arnau Ramisa
TI Learning Metrics from Teachers: Compact Networks for Image Embedding
BT 32nd IEEE Conference on Computer Vision and Pattern Recognition
PY 2019
BP 2907
EP 2916
DI 10.1109/CVPR.2019.00302
AB 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 thecommunication 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 beused on mobile devices, show that the proposed method can greatly improve Recall@1 results from 27.5% to 44.6%. Furthermore, we investigate various aspects of distillation for embeddings, including hint and attention layers, semisupervised learning and cross quality distillation.
ER