Spatio-Temporal association rule based deep annotation-free clustering (STAR-DAC) for unsupervised person re-identification

作者：

Highlights：

• We propose an unsupervised deep learning framework to label the person re-ID images.

• The proposed framework is built without any external annotations sup-port.

• We propose a spatio-temporal association rule based cluster fine-tuning.

• Proposed STAR fine-tune algorithm reduce the sample loss by 30%.

• We outperform the state-of-the-art methods in case of large multi-camera datasets.

摘要

•We propose an unsupervised deep learning framework to label the person re-ID images.•The proposed framework is built without any external annotations sup-port.•We propose a spatio-temporal association rule based cluster fine-tuning.•Proposed STAR fine-tune algorithm reduce the sample loss by 30%.•We outperform the state-of-the-art methods in case of large multi-camera datasets.

论文关键词：Unsupervised person re-identification,Clustering,Labeling,Spatio-temporal,Deep learning

论文评审过程：Received 22 August 2019, Revised 23 August 2021, Accepted 29 August 2021, Available online 30 August 2021, Version of Record 6 September 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.108287