PETAzs, RAPv2zs
收藏资源简介:
PETAzs和RAPv2zs是基于PETA和RAPv2数据集构建的两个现实场景数据集,旨在解决现有数据集中训练集和测试集身份重叠的问题。这两个数据集遵循零样本设置,确保测试集中的行人身份与训练集完全不重叠,从而更真实地反映实际应用场景。数据集的创建过程涉及对原始数据集的重新划分,以消除身份重叠,提高数据集的实用性和评估的准确性。这些数据集主要应用于行人属性识别领域,旨在解决行人属性识别模型在实际应用中的性能评估问题。
PETAzs and RAPv2zs are two real-world datasets constructed based on the PETA and RAPv2 datasets, designed to resolve the identity overlap issue between training and test sets in existing datasets. These two datasets adopt the zero-shot setting, ensuring that all pedestrian identities in the test set have no overlap with those in the training set, thus more realistically reflecting real-world application scenarios. The creation of these datasets entails re-partitioning the original datasets to eliminate identity overlap, thereby enhancing the practicality of the datasets and the accuracy of evaluation. These datasets are primarily applied in the field of pedestrian attribute recognition, aiming to solve the performance evaluation problems of pedestrian attribute recognition models in practical applications.




