Auto-ReID: Searching for a Part-aware ConvNet for Person Re-Identification
Abstract
因为re-ID 和分类任务的不同,所以会基于ResNet or VGG做修改
We propose to search for a CNN architecture that is specifically suitable for the reID task.
提出了适用于reID任务的搜索CNN架构
三个主要的问题
Thereare three main problems.
First, body structural information plays an important role in reID but is not encoded in backbones.
在backbones里没有人体架构信息的编码
Second, Neural Architecture Search (NAS) auto-mates the process of architecture design without human effort, but no existing NAS methods incorporate the structure information of input images.
现有的NAS方法并未包含输入图像的结构信息。
Third, reID is essentially a retrieval task but current NAS algorithms are merely de-signed for classification.
reID 是检索任务,但是NAS主要是为分类任务设计的
为了解决这些问题
To solve these problems, we pro-pose a retrieval-based search algorithm over a specifically designed reID search space, named Auto-ReID.
在特殊设计的reID搜索空间上的基于检索的搜索算法
Auto-ReID
Our Auto-ReID enables the automated approach to find an efficient and effective CNN architecture that is specifically suitable for reID.
Auto-ReID 实现了一种 特定适应于 reID的 高效且有效的CNN架构的自动搜索方法。
实验结果
Extensive experiments indicate that the searchedarchitecture achieves state-of-the-art performance while requiring less than about 50% parameters and 53% FLOPscompared to others.
大量实验表明,搜索结构可以实现最先进的性能,同时重新获得少于约50%的参数和53%的FLOP。