JaeYungLee / MM_GCN

PyTorch implementation of "Multi-hop Modulated Graph Convolutional Networks for 3D Human Pose Estimation", BMVC 2022

Multi-hop Modulated Graph Convolutional Networks for 3D Human Pose Estimation (BMVC2022)

Introduction

This repository holds the Pytorch implementation of Multi-hop Modulated Graph Convolutional Networks for 3D Human Pose Estimation by Jae Yung Lee and I Gil Kim.

Quick Start

This repository is build upon Python v3.6 and Pytorch v1.8.2 on Ubuntu 18.04. All experiments are conducted on a single NVIDIA RTX QUADRO 6000 GPU. See requirements.txt for other dependencies. We recommend installing Python v3.6 from Anaconda and installing Pytorch (>= 1.8.0) following guide on the official instructions according to your specific CUDA version.

Dataset

2D detections for Human3.6M datasets are provided by VideoPose3D Pavllo et al.

Pre-trained models

The pre-trained models for 3-hops can be downloaded from Google Drive.

Evaluation (GT)

Human3.6M Dataset
python test.py -d Human36M -k gt -sk {HOP_NUM} -c ${CHECKPOINT_PATH} --test_model {MODEL_PATH} -ch {CHANNEL_NUM} -j_out 17 -g {GPU_IDX}

Reference

@inproceedings{lee22multi,
 author = {Jae Yung Lee and I Gil Kim},
 booktitle = {Proceedings of the British Machine Vision Conference ({BMVC})},
 title = {Multi-hop Modulated Graph Convolutional Networks for 3D Human Pose Estimation},
 year = {2022}
}

Acknowledgement

Part of our code is borrowed from the following repositories.

• Modulated GCN
• VideoPose3D

We thank to the authors for releasing their codes. Please also consider citing their works.