ProxyNCA++: Revisiting and Revitalizing Proxy Neighborhood Component Analysis

ECCV 2020  ·  Eu Wern Teh, Terrance DeVries, Graham W. Taylor ·

We consider the problem of distance metric learning (DML), where the task is to learn an effective similarity measure between images. We revisit ProxyNCA and incorporate several enhancements... We find that low temperature scaling is a performance-critical component and explain why it works. Besides, we also discover that Global Max Pooling works better in general when compared to Global Average Pooling. Additionally, our proposed fast moving proxies also addresses small gradient issue of proxies, and this component synergizes well with low temperature scaling and Global Max Pooling. Our enhanced model, called ProxyNCA++, achieves a 22.9 percentage point average improvement of Recall@1 across four different zero-shot retrieval datasets compared to the original ProxyNCA algorithm. Furthermore, we achieve state-of-the-art results on the CUB200, Cars196, Sop, and InShop datasets, achieving Recall@1 scores of 72.2, 90.1, 81.4, and 90.9, respectively. read more

PDF Abstract ECCV 2020 PDF ECCV 2020 Abstract

Results from the Paper

 Ranked #1 on Metric Learning on Stanford Online Products (using extra training data)

     Get a GitHub badge
Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Image Retrieval CARS196 ProxyNCA++ R@1 90.1 # 2
Metric Learning CARS196 ProxyNCA++ R@1 90.1 # 2
Metric Learning CUB-200-2011 ProxyNCA++ R@1 72.2 # 2
Image Retrieval CUB-200-2011 ProxyNCA++ R@1 72.2 # 2
Image Retrieval In-Shop ProxyNCA++ R@1 90.9 # 3
Metric Learning In-Shop ProxyNCA++ R@1 90.9 # 2
Image Retrieval SOP ProxyNCA++ R@1 81.4 # 2
Metric Learning Stanford Online Products ProxyNCA++ R@1 81.4 # 1