Kali-Hac / MG-SCR

[IJCAI-2021] Official Codes for "Multi-Level Graph Encoding with Structural-Collaborative Relation Learning for Skeleton-Based Person Re-Identification"

Multi-Level Graph Encoding with Structural-Collaborative Relation Learning for Skeleton-Based Person Re-Identification

By Haocong Rao, Shihao Xu, Xiping Hu, Jun Cheng, Bin Hu. In IJCAI 2021.

Introduction

This is the official implementation of MG-SCR model presented by "Multi-Level Graph Encoding with Structural-Collaborative Relation Learning for Skeleton-Based Person Re-Identification". The codes are used to reproduce experimental results in the paper.

Abstract: Skeleton-based person re-identification (Re-ID) is an emerging open topic providing great value for safety-critical applications. Existing methods typically extract hand-crafted features or model skeleton dynamics from the trajectory of body joints, while they rarely explore valuable relation information contained in body structure or motion. To fully explore body relations, we construct graphs to model human skeletons from different levels, and for the first time propose a Multi-level Graph encoding approach with Structural-Collaborative Relation learning (MG-SCR) to encode discriminative graph features for person Re-ID. Specifically, considering that structurally-connected body components are highly correlated in a skeleton, we first propose a multi-head structural relation layer to learn different relations of neighbor body-component nodes in graphs, which helps aggregate key correlative features for effective node representations. Second, inspired by the fact that body-component collaboration in walking usually carries recognizable patterns, we propose a cross-level collaborative relation layer to infer collaboration between different level components, so as to capture more discriminative skeleton graph features. Finally, to enhance graph dynamics encoding, we propose a novel self-supervised sparse sequential prediction task for model pre-training, which facilitates encoding high-level graph semantics for person Re-ID. MG-SCR outperforms state-of-the-art skeleton-based methods, and it can achieve superior performance to many multi-modal methods that utilize extra RGB or depth information.

Requirements

• Python 3.5
• Tensorflow 1.10.0 (GPU)

Datasets and Models

We provide three already pre-processed datasets (BIWI, IAS, KGBD) with various sequence lengths on
https://pan.baidu.com/s/1fHMbYZ8H3hZYqrtENupBmA       password：  jtof

All the best models reported in our paper can be acquired on
https://pan.baidu.com/s/1EhPWg6pJ0Vl4xOh0swQSDQ       password：  e7oq
Please download the pre-processed datasets Datasets/ and model files trained_models/ into the current directory.

The original datasets can be downloaded here: BIWI and IAS-Lab, KGBD, KS20.

Note: The access to the Vislab Multi-view KS20 dataset is available upon request. If you have signed the license agreement and been granted the right to use it, please contact me and I will share the pre-processed KS20 data.

Usage

To (1) pre-train the MG-SCR model by sparse sequential prediction (SSP) and (2) fine-tune the model for person Re-ID on a specific dataset, simply run the following command:

python train.py --dataset BIWI

# Default options: --dataset BIWI --split '' --length 6 --c_lambda 0.3 --task pre --gpu 0
# --dataset [BIWI, IAS, KGBD, KS20]
# --split ['A' (for IAS-A), 'B' (for IAS-B)] 
# --length [4, 6, 8, 10] 
# --task ['pre' (use SSP pre-training), 'none' (no pre-training)]
# --gpu [0, 1, ...]

Please see train.py for more details.

To print evaluation results (Re-ID Confusion Matrix / Rank-n Accuracy / Rank-1 Accuracy / nAUC) of the trained model, run:

python evaluate.py --dataset BIWI --model_dir trained_models/xx/xx

# Default options: --dataset BIWI --model_dir best --length 6 --gpu 0
# --dataset [BIWI, IAS, KGBD, KS20] 
# --model_dir [best (load the best models), trained_models/xx/xx (directory of model files, e.g., trained_models/best_models/KS20_87.3_95.5)] 

Please see evaluate.py for more details.

Application to Model-Estimated Skeleton Data

To extend our model to a large RGB-based gait dataset (CASIA B), we exploit pose estimation methods to extract 3D skeletons from RGB videos of CASIA B as follows:

• Step 1: Download CASIA-B Dataset
• Step 2: Extract the 2D human body joints by using OpenPose
• Step 3: Estimate the 3D human body joints by using 3DHumanPose

We provide already pre-processed skeleton data of CASIA B for Cross-View Evaluation (CVE) (f=20/30/40) on       https://pan.baidu.com/s/1gDekBzf-3bBVdd0MGL0CvA       password：  x3e5
Please download the pre-processed datasets into the directory Datasets/.

Usage

To (1) pre-train the MG-SCR model by sparse sequential prediction (SSP) and (2) fine-tune the model for person Re-ID on CASIA B under CVE setup, simply run the following command:

python train-CASIA.py --view 0

# Default options: --dataset CASIA_B --split '' --length 20 --c_lambda 0.3 --task pre --gpu 0
# --length [20, 30, 40] 
# --view [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10] 
# --task ['pre' (use SSP pre-training), 'none' (no pre-training)]
# --gpu [0, 1, ...]

Please see train-CASIA.py for more details.

Results

Models	Rank-1					nAUC
	BIWI	IAS-A	IAS-B	KS20	KGBD	BIWI	IAS-A	IAS-B	KS20	KGBD
Gait Energy Image	21.4	25.6	15.9	—	—	73.2	72.1	66.0	—	—
3D CNN + Average Pooling	27.8	33.4	39.1	—	—	84.0	81.4	82.8	—	—
Gait Energy Volume	25.7	20.4	13.7	—	—	83.2	66.2	64.8	—	—
3D LSTM	27.0	31.0	33.8	—	—	83.3	77.6	78.0	—	—
Size-Shape Descriptors + SVM	20.5	—	—	—	—	87.2	—	—	—	—
Size-Shape Descriptors + LDA	22.1	—	—	—	—	88.5	—	—	—	—
DVCov + SKL	21.4	46.6	45.9	—	—	—	—	—	—	—
ED + SKL	30.0	52.3	63.3	—	—	—	—	—	—	—
CNN-LSTM with RTA	50.0	—	—	—	—	—	—	—	—	—
D-13 Descriptors + KNN	39.3	33.8	40.5	58.3	46.9	64.3	63.6	71.1	78.0	90.0
Single-layer LSTM	15.8	20.0	19.1	80.9	39.8	65.8	65.9	68.4	92.3	87.2
Multi-layer LSTM	36.1	34.4	30.9	81.6	46.2	75.6	72.1	71.9	94.2	89.8
D-16 Descriptors + Adaboost	41.8	27.4	39.2	59.8	69.9	74.1	65.5	78.2	78.8	90.6
PoseGait	33.3	41.4	37.1	70.5	90.6	81.8	79.9	74.8	94.0	97.8
Attention Gait Encodings	59.1	56.1	58.2	86.9	87.7	86.5	81.7	85.3	94.7	96.3
MG-SCR (Ours)	61.6	56.5	65.9	87.3	96.3	91.9	87.0	93.1	95.5	99.9

Model Size & Computational Complexity

Methods	# Params	GFLOPs
PoseGait	26.80M	364.8
AGE	7.15M	37.4
CAGE	8.47M	7.5
MG-SCR (Ours)	1.04M	19.8

Citation

If you find this code useful for your research, please cite our paper

@article{rao2021multi,
  title={Multi-Level Graph Encoding with Structural-Collaborative Relation Learning for Skeleton-Based Person Re-Identification},
  author={Rao, Haocong and Xu, Shihao and Hu, Xiping and Cheng, Jun and Hu, Bin},
  journal={arXiv preprint arXiv:2106.03069},
  year={2021}
}
@inproceedings{ijcai2021-0135,
  title     = {Multi-Level Graph Encoding with Structural-Collaborative Relation Learning for Skeleton-Based Person Re-Identification},
  author    = {Rao, Haocong and Xu, Shihao and Hu, Xiping and Cheng, Jun and Hu, Bin},
  booktitle = {Proceedings of the Thirtieth International Joint Conference on
               Artificial Intelligence, {IJCAI-21}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  editor    = {Zhi-Hua Zhou},
  pages     = {973--980},
  year      = {2021},
  month     = {8},
  note      = {Main Track}
  doi       = {10.24963/ijcai.2021/135},
  url       = {https://doi.org/10.24963/ijcai.2021/135},
}

More awesome skeleton-based models are collected in our Awesome-Skeleton-Based-Models.

Acknowledgements

Thanks to Veličković et al. for opening source of their excellent works GAT.

License

MG-SCR is released under the MIT License.