Unsupervised domain adaptation pedestrian re-identification based on an improved dissimilarity space

作者：

Highlights：

• The test results show that our method can improve the performance of cross-domain Re-ID effectively.

• Based on the DMMD model proposed in ECCV2020, we propose an improved dissimilarity space.

• For transfer learning, we propose a new classification boundary optimization method.

• For feature representation, we propose a time-space-appearance loss.

摘要

Highlights•The test results show that our method can improve the performance of cross-domain Re-ID effectively.•Based on the DMMD model proposed in ECCV2020, we propose an improved dissimilarity space.•For transfer learning, we propose a new classification boundary optimization method.•For feature representation, we propose a time-space-appearance loss.

论文关键词：Transfer learning,Cross-domain,Pedestrian re-identification,Maximum mean discrepancy,Dissimilarity space

论文评审过程：Received 1 August 2021, Revised 22 November 2021, Accepted 7 December 2021, Available online 14 December 2021, Version of Record 3 January 2022.

论文官网地址：https://doi.org/10.1016/j.imavis.2021.104354