Learning Generalisable Omni-Scale Representations for Person Re-Identification

TL;DR

This paper introduces OSNet, a lightweight CNN architecture with omni-scale features and instance normalization, achieving state-of-the-art person re-identification performance and strong cross-dataset generalization without domain adaptation.

Contribution

The paper proposes a novel omni-scale CNN architecture with dynamic multi-scale feature fusion and an architecture search for optimal IN layer placement, enhancing generalization in person re-ID.

Findings

OSNet achieves state-of-the-art results on standard datasets.

OSNet outperforms many unsupervised domain adaptation methods cross-dataset.

The architecture is lightweight and efficient.

Abstract

An effective person re-identification (re-ID) model should learn feature representations that are both discriminative, for distinguishing similar-looking people, and generalisable, for deployment across datasets without any adaptation. In this paper, we develop novel CNN architectures to address both challenges. First, we present a re-ID CNN termed omni-scale network (OSNet) to learn features that not only capture different spatial scales but also encapsulate a synergistic combination of multiple scales, namely omni-scale features. The basic building block consists of multiple convolutional streams, each detecting features at a certain scale. For omni-scale feature learning, a unified aggregation gate is introduced to dynamically fuse multi-scale features with channel-wise weights. OSNet is lightweight as its building blocks comprise factorised convolutions. Second, to improve generalisable feature learning, we introduce instance normalisation (IN) layers into OSNet to cope with cross-dataset discrepancies. Further, to determine the optimal placements of these IN layers in the architecture, we formulate an efficient differentiable architecture search algorithm. Extensive experiments show that, in the conventional same-dataset setting, OSNet achieves state-of-the-art performance, despite being much smaller than existing re-ID models. In the more challenging yet practical cross-dataset setting, OSNet beats most recent unsupervised domain adaptation methods without using any target data. Our code and models are released at \texttt{https://github.com/KaiyangZhou/deep-person-reid}.