Re-ID done right: towards good practices for person re-identification

Almazan J, Gajic B, Murray N, et al. Re-ID done right: towards good practices for person re-identification[J]. 2018.

  这篇总结了一套行之有效的Re-id训练策略和调参经验。创新点不多,但对工程有较好的借鉴意义。

contributions

1.总结了一组关键措施和有效的训练策略,能捕获差异特征
2.利用特征可视化进行论证分析

Good Practices

因素1:back bone

可以使用ResNet,ResNeXt,Inception,Densenet等。
使用pretrained训练性能较好。

Person Re-identification 系列论文笔记(四):Re-ID done right: towards good practices for person re-identification

因素2:图像分辨率

分辨率较大性能较好,但有瓶颈。

Person Re-identification 系列论文笔记(四):Re-ID done right: towards good practices for person re-identification

因素3:样本增强

cut-out(在局部添加噪声,类似dropout的操作)能有效增强鲁棒和降低过拟合。

Person Re-identification 系列论文笔记(四):Re-ID done right: towards good practices for person re-identification

Person Re-identification 系列论文笔记(四):Re-ID done right: towards good practices for person re-identification

因素4:Hard Triplet Mining

triplet中添加困难样本挖掘,能显著提高性能

 Person Re-identification 系列论文笔记(四):Re-ID done right: towards good practices for person re-identification

特征可视化结果:

捕捉到了全局和局部特征

Person Re-identification 系列论文笔记(四):Re-ID done right: towards good practices for person re-identification

 

相关文章:

  • 2021-11-24
  • 2018-07-26
  • 2022-12-23
  • 2021-10-25
  • 2021-08-30
  • 2022-01-20
  • 2021-05-14
猜你喜欢
  • 2021-05-08
  • 2022-12-23
  • 2022-12-23
  • 2021-06-05
  • 2022-12-23
  • 2021-05-06
  • 2021-11-28
相关资源
相似解决方案