Auto-ReID: Searching for a Part-aware ConvNet for Person Re-Identification

TL;DR

Auto-ReID introduces a neural architecture search method tailored for person re-identification, focusing on incorporating body structure information and optimizing CNNs specifically for retrieval tasks, leading to state-of-the-art results.

Contribution

The paper presents Auto-ReID, a novel NAS framework that automatically designs CNN architectures optimized for person reID, integrating structural information and retrieval-specific considerations.

Findings

Achieves state-of-the-art reID performance

Reduces 50% parameters compared to existing models

Reduces 53% FLOPs, improving efficiency

Abstract

Prevailing deep convolutional neural networks (CNNs) for person re-IDentification (reID) are usually built upon ResNet or VGG backbones, which were originally designed for classification. Because reID is different from classification, the architecture should be modified accordingly. We propose to automatically search for a CNN architecture that is specifically suitable for the reID task. There are three aspects to be tackled. First, body structural information plays an important role in reID but it is not encoded in backbones. Second, Neural Architecture Search (NAS) automates the process of architecture design without human effort, but no existing NAS methods incorporate the structure information of input images. Third, reID is essentially a retrieval task but current NAS algorithms are merely designed for classification. To solve these problems, we propose a retrieval-based search algorithm over a specifically designed reID search space, named Auto-ReID. Our Auto-ReID enables the automated approach to find an efficient and effective CNN architecture for reID. Extensive experiments demonstrate that the searched architecture achieves state-of-the-art performance while reducing 50% parameters and 53% FLOPs compared to others.