Self-Supervised Gait Encoding with Locality-Aware Attention for Person Re-Identification

By Haocong Rao, Siqi Wang, Xiping Hu, Mingkui Tan, Huang Da, Jun Cheng, Bin Hu. In IJCAI 2020.

Introduction

This is the official implementation of the self-supervised gait encoding model presented by "Self-Supervised Gait Encoding with Locality-Aware Attention for Person Re-Identification". The codes are used to reproduce experimental results of the proposed Attention-basd Gait Encodings (AGEs) in the paper.

Abstract: Gait-based person re-identification (Re-ID) is valuable for safety-critical applications, and using only 3D skeleton data to extract discriminative gait features for person Re-ID is an emerging open topic. Existing methods either adopt hand-crafted features or learn gait features by traditional supervised learning paradigms. Unlike previous methods, we for the first time propose a generic gait encoding approach that can utilize unlabeled skeleton data to learn gait representations in a self-supervised manner. Specifically, we first propose to introduce self-supervision by learning to reconstruct input skeleton sequences in reverse order, which facilitates learning richer high-level semantics and better gait representations. Second, inspired by the fact that motion's continuity endows temporally adjacent skeletons with higher correlations (“locality”), we propose a locality-aware attention mechanism that encourages learning larger attention weights for temporally adjacent skeletons when reconstructing current skeleton, so as to learn locality when encoding gait. Finally, we propose Attention-based Gait Encodings (AGEs), which are built using context vectors learned by locality-aware attention, as final gait representations. AGEs are directly utilized to realize effective person Re-ID. Our approach typically improves existing skeleton-based methods by 10-20% Rank-1 accuracy, and it achieves comparable or even superior performance to multi-modal methods with extra RGB or depth information.

Requirements

Python 3.5
Tensorflow 1.10.0 (GPU)

Datasets & Models

We provide three already preprocessed datasets (BIWI, IAS, KGBD) on
Google Cloud Baidu Cloud Password: kle5 Tencent Cloud password： ma385h

Two already trained models (BIWI, IAS) are saved in this repository, and all three models can be acquired on
Google Cloud Baidu Cloud Password: r1jp Tencent Cloud password： 6xpj8r
Please download the preprocessed datasets Datasets/ and the model files Models/ into the current directory.

The original datasets can be downloaded from: BIWI and IAS-Lab KGBD

Usage

To (1) train the self-supervised gait encoding model to obtain AGEs and (2) validate the effectiveness of AGEs for person Re-ID on a specific dataset with a recognition network, simply run the following command:

# --attention: LA (default), BA  --dataset: BIWI, IAS, KGBD  --gpu 0 (default)
python train.py --dataset BIWI

Please see train.py for more details.

To print evaluation results (Rank-1 accuracy/nAUC) of person re-identification (Re-ID) on the testing set, run:

# --attention: LA (default), BA  --dataset: BIWI, IAS, KGBD  --gpu 0 (default)
python evaluate.py --dataset BIWI

Please see evaluate.py for more details.

Citation

@inproceedings{rao2020self,
	title="Self-Supervised Gait Encoding with Locality-Aware Attention for Person Re-Identification",
	author="Haocong {Rao} and Siqi {Wang} and Xiping {Hu} and Mingkui {Tan} and Huang {Da} and Jun {Cheng} and Bin {Hu}",
	booktitle="IJCAI 2020: International Joint Conference on Artificial Intelligence",
	year="2020"
}

License

SGE-LA is released under the MIT License.

SCUT-AILab / SGE-LA