采样(sample)

一、硬采样(hard-sample)

1、hard-positive-sample

错误分类为负类的正类样本。

2、hard-negative-sample

错误分类为正类的负类样本。

http://t.csdnimg.cn/guSEe

二、半硬采样(semi-hard-sample)

http://t.csdnimg.cn/vOaES

FaceNet: A Unified Embedding for Face Recognition and Clustering

采用Semi-Hard Negative Sampling

Schroff, Florian, et al. “FaceNet: A Unified Embedding for Face Recognition and Clustering.” 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015, https://doi.org/10.1109/cvpr.2015.7298682.

Sampling Matters in Deep Embedding Learning

采用Distance Weighted Sampling 和 Semi-Hard Negative Sampling

三、智能挖掘(Smart Mining)

Smart Mining for Deep Metric Learning

Harwood, Ben, et al. “Smart Mining for Deep Metric Learning.” 2017 IEEE International Conference on Computer Vision (ICCV), 2017, https://doi.org/10.1109/iccv.2017.307.(Smart mining)