Existing gait recognition frameworks retrieve an identity in the gallery based on the distance between a probe sample and the identities in the gallery. However, existing methods often neglect that the gallery may not contain identities corresponding to the probes, leading to recognition errors rather than raising an alarm. In this paper, we introduce a novel uncertainty-aware gait recognition method that models the uncertainty of identification based on learned evidence. Specifically, we treat our recognition model as an evidence collector to gather evidence from input samples and parameterize a Dirichlet distribution over the evidence. The Dirichlet distribution essentially represents the density of the probability assigned to the input samples. We utilize the distribution to evaluate the resultant uncertainty of each probe sample and then determine whether a probe has a counterpart in the gallery or not. To the best of our knowledge, our method is the first attempt to tackle gait recognition with uncertainty modelling. Moreover, our uncertain modeling significantly improves the robustness against out-of-distribution (OOD) queries. Extensive experiments demonstrate that our method achieves state-of-the-art performance on datasets with OOD queries, and can also generalize well to other identity-retrieval tasks. Importantly, our method outperforms the state-of-the-art by a large margin of 51.26% when the OOD query rate is around 50% on OUMVLP.

翻译：现有步态识别框架基于探针样本与库中身份之间的距离检索对应身份。然而，现有方法常忽略库中可能不含与探针匹配的身份，导致识别错误而非触发告警。本文提出一种新颖的不确定性感知步态识别方法，基于学习证据建模识别的不确定性。具体而言，我们将识别模型视为证据收集器，从输入样本中收集证据，并基于证据参数化狄利克雷分布。该分布本质上表征了输入样本概率分配的密度。我们利用该分布评估每个探针样本的结果不确定性，进而判断探针在库中是否存在对应身份。据我们所知，本方法首次尝试将不确定性建模应用于步态识别。此外，我们的不确定性建模显著提升了对外分布查询的鲁棒性。大量实验表明，本方法在含外分布查询的数据集上达到最优性能，并能良好泛化至其他身份检索任务。重要的是，在OUMVLP数据集中当外分布查询率约为50%时，本方法性能以51.26%的绝对优势超越现有最优方法。