Loss function search for person re-identification

In recent years, person re-identification, which learns discriminative features for the specific person retrieval problem across non-overlapping cameras, has attracted extensive attention. One of the main challenges in person re-identification with deep neural networks is the design of the loss function, which plays a vital role in improving the discrimination of the learned features. However, most existing models utilize the hand-designed loss functions, which are usually sub-optimal and time-consuming. The search spaces of the two existing AutoML-based methods are either too complicated or too simple to include various forms of loss functions. In order to solve the irrationality of the above search spaces, in this paper, we propose a method of AutoML for loss function search named LFS-ReID for person ReID in the framework of the margin-based softmax loss function. Specifically, we first analyze the margin-based softmax loss function and conclude four key properties. Then we carefully design a sampling distribution based on the non-independent truncated Gaussian distributions to sample the loss function, which conforms to the above four properties. Finally, a method based on reinforcement learning is adopted to optimize the sampling distribution dynamically. Experimental results demonstrate that the proposed method achieves state-of-the-art performance on four commonly used datasets.

Automated summary

Person re-identification, a technique that learns discriminative features for person retrieval across non-overlapping cameras, has gained significant attention in recent years. The design of loss function plays a crucial role in improving the discrimination of the learned features. However, existing models mostly use sub-optimal and time-consuming hand-designed loss functions. This paper proposes an AutoML-based method, named LFS-ReID, to optimize the loss function search for person re-identification using the margin-based softmax loss function. Experimental results demonstrate that the proposed method achieves state-of-the-art performance on commonly used datasets.