深度學習re-id綜述

related work

• 多尺度網絡 參數過多
  1.另外加局部特徵分支
  

2. 使用多尺度的卷積流
   OSnet

• 多層級

1. 使用附加的屬性信息
   
2. 融合多層信息
   
3. deep supervision
   

特徵提取

global feature

• attention

local feature

人體部分/區域特徵的集合，解決對齊問題，身體部分由人體姿態構建或者粗略的垂直分割完成

Auxilary 輔助方法

需要額外的標註信息

• 語意屬性標註
• view
• camera
• data augumentation

1. 位姿約束
   
   guild 指導符合原來的分類

2. 相機風格信息
   
   camstyle

3. 將外觀和結構分離生成圖片
   

4. 無監督域適應
   
   SPGAN eSPGAN

Person Transfer GAN to Bridge Domain Gap for Person Re-Identification- PTGAN
Wei et al. [35] handle the domain
gap problem by proposing a Person Transfer Generative
Adversarial Network (PTGAN), transferring the knowledge
from one labeled source dataset to another unlabeled target
dataset.
An image-image domain adaptation method [118]
is developed with preserved self-similarity and domaindissimilarity, trained with a similarity preserving generative adversarial network (SPGAN).
A Hetero-Homogeneous Learning (HHL) method [215] simultaneously considers the
camera invariance with homogeneous learning (image pairs
from the same domain) and domain connectedness with
heterogeneous learning (cross-domain negative pairs).

• 網絡結構
  OmniScale Network (OSNet) 多尺度

Multi-Level Factorisation Net (MLFN) 多層級

度量函數

• identity loss 分類損失
  
• varification loss
  區分這一對圖片是否是同一個人
  
• triplet loss
  

優化方法

re-ranking 使用gallery-to-gallery 的相似性來優化初始序列。

待解決問題

Scalable Re-ID

• Lightweight Model. 更輕型的結構、模型蒸餾
• Resource Aware Re-ID.根據硬件配置適應性地調整模型 Deep Anytime ReID (DaRe)

Domain-Specific Architecture Design

多基於分類網絡，爲re-id網絡提出新的結構