FedReIDBench
收藏资源简介:
FedReIDBench是一个包含9个不同规模和来源的数据集的基准,用于模拟现实世界中的统计异质性情况。这些数据集包括MSMT17、DukeMTMC-reID、Market-1501、CUHK03-NP、PRID2011、CUHK01、VIPeR、3DPeS和iLIDS-VID,用于实现联邦学习到人物再识别(FedReID)。数据集的创建旨在通过模拟数据点的不平衡和非独立同分布(non-IID)问题,来研究FedReID在真实世界场景中的性能优化。该基准还定义了代表性的联邦学习场景,并提出了适用于FedReID的算法,以及标准化模型结构和性能评估指标,为未来的研究和工业化提供了有价值的见解和基准。
FedReIDBench is a benchmark comprising nine datasets with varying scales and sources, designed to simulate statistical heterogeneity in real-world scenarios. These datasets include MSMT17, DukeMTMC-reID, Market-1501, CUHK03-NP, PRID2011, CUHK01, VIPeR, 3DPeS and iLIDS-VID, which are employed to implement Federated Person Re-identification (FedReID). The benchmark is developed to study the performance optimization of FedReID in real-world scenarios by simulating data imbalance and non-independent and identically distributed (non-IID) problems. Additionally, this benchmark defines representative federated learning scenarios, proposes algorithms tailored for FedReID, as well as standardized model structures and performance evaluation metrics, providing valuable insights and benchmarks for future research and industrial applications.

- 1Performance Optimization for Federated Person Re-identification via Benchmark Analysis南洋理工大学, 新加坡 · 2020年



