一、前面

最近入门 Person Re-id，用老办法，一篇一篇文献地追读，下面记录一些文献里面有意思的话。

二、2015相关paper

1. Ranking optimization for person re-identification via similarity and dissimilarity.

• we address ranking optimization approaches for person re-identification problem, a novel and efficient
  ranking optimization method based on similarity and dissimilarity is conducted. The main idea is that the correct match should be similar to the probe’s strongly similar samples and dissimilar to the strongly dissimilar samples, we pull the quasi-similar samples and push the quasi-dissimilar samples to optimize the ranking list.

  
  main idea of this paper

这篇论文最大的贡献就是：既取top-10作为probe的相似，又取bottom-10作为probe的极度不相似。相似和极度不相似来更新rank。

• Person re-identification is a key technique to match different persons observed in non-overlapping camera views.

• existing person re-id methods can be roughly divided into three categories：

  1. appearance representation methods：aim to seek distinctive descriptions by utilizing the structure and position of the human body.
  2. metric learning methods： attempt to learn a proper distance metric to obtain an accurate similarity measurement to reduce the differences in disjoint camera images.
  3. ranking optimization methods：try to refine the original ranking lists by establishing the similarity relationships lying in the identification results.

• While the ranking optimization methods based on ranking orders do not rely on the similarity scores, which own more flexibility and popularity, thus can be easily reproduced.
• Related ranking optimization methods can be categorized as automotive re-ranking and interactive relevance feedback.