Re-ID 2019 Review

Academic Report

Posted by JoselynZhao on October 19, 2019


Train/Test on the same domain

Part-level features are effective. 在这里插入图片描述 pose and 3D information are beneficial 在这里插入图片描述

**Smothness in feature space is beneficial ** 在这里插入图片描述

Unsupervised domain adaptation

Evolution in state-of-the art Performance 在这里插入图片描述

image-image translation benefits UDA; identity-preserving property benefits further 在这里插入图片描述

mining discriminative cues in the target domain improves the UDA accuracy

挖掘 区别性大的线索 提高 unsupervised domain adaptation accuracy.

Video re-id: frame-level weights are important 在这里插入图片描述 spatiotemporal attention network 在这里插入图片描述


person search

end-to-end framework 在这里插入图片描述

feature memory modeling is critical for end to end person search


separate person detector and re-identification seem to be a better choice.


Other problems

person re-id originates from person tracking; After years of study, person re-id features are improving tracking accuracies. 在这里插入图片描述


learning from synthetic data for person re-id **Synthetic training data can help to initialize deep networds **

在这里插入图片描述 the diversity of synthetic data can help improve the generalization performance of re-ID models

合成数据的多样性…… 在这里插入图片描述 在这里插入图片描述


Future research questions

  1. is large-scale re-id really solved? what the performance will be if we scale up the gallery to 1 million or 10 million images? How to accelerate?
  2. In unsupervised domain adaptation, person re-id is a open-set problem. what is its relationship with UDA in image classification?
  3. What is the relationship between person detection and re-id? is it optimal to use re-id features in tracking?
  4. How do environmental factors affect re-id?