Graph Stacked Hourglass Network (CVPR 2021)

This repository contains the pytorch implementation of the approach described in the paper:

Tianhan Xu and Wataru Takano. Graph Stacked Hourglass Networks for 3D Human Pose Estimation Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021, pp. 16105-16114

@InProceedings{Xu_2021_CVPR,
    author    = {Xu, Tianhan and Takano, Wataru},
    title     = {Graph Stacked Hourglass Networks for 3D Human Pose Estimation},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2021},
    pages     = {16105-16114}
}

Introduction

We evaluate models for 3D human pose estimation on the Human3.6M Dataset.

In this repository, only 2D joints of the human pose are exploited as inputs. We utilize the method described in Pavllo et al. [2] to normalize 2D and 3D poses in the dataset. To be specific, 2D poses are scaled according to the image resolution and normalized to [-1, 1]; 3D poses are aligned with respect to the root joint. Please refer to the corresponding part in Pavllo et al. [2] for more details. For the 2D ground truth, we predict 16 joints (as the skeleton in Martinez et al. [1] and Zhao et al. [3] without the 'Neck/Nose' joint). For the 2D pose detections, the 'Neck/Nose' joint is reserved.

Quickstart

This repository is build upon Python v3.8 and Pytorch v1.9.0 on Ubuntu 20.04 LTS. All experiments are conducted on a single NVIDIA RTX 3090 GPU. See requirements.txt for other dependencies. We recommend installing Python v3.78 from Anaconda and installing Pytorch (>= 1.9.0) following guide on the official instructions according to your specific CUDA version. Then you can install dependencies with the following commands.

git clone https://github.com/tamasino52/GraphSH.git
cd GraphSH
pip install -r requirements.txt

Benchmark setup

CPN 2D detections for Human3.6M datasets are provided by VideoPose3D Pavllo et al. [2], which can be downloaded by the following steps:

cd data
wget https://dl.fbaipublicfiles.com/video-pose-3d/data_2d_h36m_cpn_ft_h36m_dbb.npz
wget https://dl.fbaipublicfiles.com/video-pose-3d/data_2d_h36m_detectron_ft_h36m.npz
cd ..

3D labels and ground truth can be downloaded

cd data
gdown --id 1P7W3ldx2lxaYJJYcf3RG4Y9PsD4EJ6b0
cd ..

GT setup

GT 2D keypoints for Human3.6M datasets are obtained from SemGCN Zhao et al. [3], which can be downloaded by the following steps:

cd data
pip install gdown
gdown https://drive.google.com/uc?id=1Ac-gUXAg-6UiwThJVaw6yw2151Bot3L1
python prepare_data_h36m.py --from-archive h36m.zip
cd ..

After this step, you should end up with two files in the dataset directory: data_3d_h36m.npz for the 3D poses, and data_2d_h36m_gt.npz for the ground-truth 2D poses.

GT Evaluation

python main_gcn.py --gcn {gcn_name}  --evaluate checkpoint/{weight_name}.pth.tar

GT Training

# Decoupled Vanilla GCN (What method used in paper)
python main_gcn.py --gcn dc_vanilla

# Decoupled Pre-Aggresive GCN (What method used in paper)
python main_gcn.py --gcn dc_preagg

# Semantic GCN (What method used in paper)
python main_gcn.py --gcn semantic

# Decoupled Post-Aggresive GCN
python main_gcn.py --gcn dc_postagg

# Convolution-style GCN
python main_gcn.py --gcn convst

# No-sharing GCN
python main_gcn.py --gcn nosharing

# Modulated GCN
python main_gcn.py --gcn modulated

Training

# Decoupled Vanilla GCN
python main_gcn.py --gcn dc_vanilla --keypoints cpn_ft_h36m_dbb

# Decoupled Pre-Aggresive GCN
python main_gcn.py --gcn dc_preagg --keypoints cpn_ft_h36m_dbb

Pre-trained weight

I implemented and tested all the elements proposed in this paper, but failed to reach the benchmark score presented in the paper. Instead, I uploaded a good enough one, so use it if you need it.

Download Link.

Acknowledgement

This code is extended from the following repositories.

• 3d-pose-baseline
• 3d_pose_baseline_pytorch
• VideoPose3D
• Semantic GCN
• Local-to-Global GCN
• Modulated-GCN

Thank you to authors for releasing their code. Please also consider citing their works.