PoolNet deep feature based person re-identification

Learning with Deep Neural Networks has recently reached state-of-the-art outcomes for Person Re-Identification. Effective learning can be accomplished only with efficient features robust to illumination and viewpoint changes. This paper proposes a new feature representation method called PoolNet Deep Feature (PNDF) for person re-identification with Convolution Neural Networks. The proposed CNN architecture called PoolNet consists of two Pool Added Blocks (PAB) and a Pool Concatenated Block (PCB) to extract the more sophisticated dominant and precise features for better learning towards a person's re-identification. The efficiency of the proposed method is demonstrated in terms of re-identification accuracy by implementing it on the challenging small scale & large-scale person re-identification datasets such as VIPeR, Market1501, CUHK03, GRID, and LaST.

Automated summary

In this paper, a new feature representation method called PNDF is proposed for person re-identification using Convolution Neural Networks. The proposed method, based on the PoolNet CNN architecture, extracts more sophisticated and precise features for better learning. The efficiency of the proposed method is demonstrated through experiments on challenging person re-identification datasets.