Direct Multi-view Multi-person 3D Human Pose Estimation

Related tags

Miscellaneousmvp
Overview

Implementation of NeurIPS-2021 paper: Direct Multi-view Multi-person 3D Human Pose Estimation

[paper] [video-YouTube, video-Bilibili] [slides]

This is the official implementation of our NeurIPS-2021 work: Multi-view Pose Transformer (MvP). MvP is a simple algorithm that directly regresses multi-person 3D human pose from multi-view images.

Framework

mvp_framework

Example Result

mvp_framework

Reference

@article{wang2021mvp,
  title={Direct Multi-view Multi-person 3D Human Pose Estimation},
  author={Tao Wang and Jianfeng Zhang and Yujun Cai and Shuicheng Yan and Jiashi Feng},
  journal={Advances in Neural Information Processing Systems},
  year={2021}
}

1. Installation

  1. Set the project root directory as ${POSE_ROOT}.
  2. Install all the required python packages (with requirements.txt).
  3. compile deformable operation for projective attention.
cd ./models/ops
sh ./make.sh

2. Data and Pre-trained Model Preparation

2.1 CMU Panoptic

Please follow VoxelPose to download the CMU Panoptic Dataset and PoseResNet-50 pre-trained model.

The directory tree should look like this:

${POSE_ROOT}
|-- models
|   |-- pose_resnet50_panoptic.pth.tar
|-- data
|   |-- panoptic
|   |   |-- 16060224_haggling1
|   |   |   |-- hdImgs
|   |   |   |-- hdvideos
|   |   |   |-- hdPose3d_stage1_coco19
|   |   |   |-- calibration_160224_haggling1.json
|   |   |-- 160226_haggling1
|   |   |-- ...

2.2 Shelf/Campus

Please follow VoxelPose to download the Shelf/Campus Dataset.

Due to the limited and incomplete annotations of the two datasets, we use psudo ground truth 3D pose generated from VoxelPose to train the model, we expect mvp would perform much better with absolute ground truth pose data.

Please use voxelpose or other methods to generate psudo ground truth for the training set, you can also use our generated psudo GT: psudo_gt_shelf. psudo_gt_campus. psudo_gt_campus_fix_gtmorethanpred.

Due to the small dataset size, we fine-tune Panoptic pre-trained model to Shelf and Campus. Download the pretrained MvP on Panoptic from model_best_5view and model_best_3view_horizontal_view or model_best_3view_2horizon_1lookdown

The directory tree should look like this:

${POSE_ROOT}
|-- models
|   |-- model_best_5view.pth.tar
|   |-- model_best_3view_horizontal_view.pth.tar
|   |-- model_best_3view_2horizon_1lookdown.pth.tar
|-- data
|   |-- Shelf
|   |   |-- Camera0
|   |   |-- ...
|   |   |-- Camera4
|   |   |-- actorsGT.mat
|   |   |-- calibration_shelf.json
|   |   |-- pesudo_gt
|   |   |   |-- voxelpose_pesudo_gt_shelf.pickle
|   |-- CampusSeq1
|   |   |-- Camera0
|   |   |-- Camera1
|   |   |-- Camera2
|   |   |-- actorsGT.mat
|   |   |-- calibration_campus.json
|   |   |-- pesudo_gt
|   |   |   |-- voxelpose_pesudo_gt_campus.pickle
|   |   |   |-- voxelpose_pesudo_gt_campus_fix_gtmorethanpred_case.pickle

2.3 Human3.6M dataset

Please follow CHUNYUWANG/H36M-Toolbox to prepare the data.

2.4 Full Directory Tree

The data and pre-trained model directory tree should look like this, you can only download the Panoptic dataset and PoseResNet-50 for reproducing the main MvP result and ablation studies:

${POSE_ROOT}
|-- models
|   |-- pose_resnet50_panoptic.pth.tar
|   |-- model_best_5view.pth.tar
|   |-- model_best_3view_horizontal_view.pth.tar
|   |-- model_best_3view_2horizon_1lookdown.pth.tar
|-- data
|   |-- pesudo_gt
|   |   |-- voxelpose_pesudo_gt_shelf.pickle
|   |   |-- voxelpose_pesudo_gt_campus.pickle
|   |   |-- voxelpose_pesudo_gt_campus_fix_gtmorethanpred_case.pickle
|   |-- panoptic
|   |   |-- 16060224_haggling1
|   |   |   |-- hdImgs
|   |   |   |-- hdvideos
|   |   |   |-- hdPose3d_stage1_coco19
|   |   |   |-- calibration_160224_haggling1.json
|   |   |-- 160226_haggling1
|   |   |-- ...
|   |-- Shelf
|   |   |-- Camera0
|   |   |-- ...
|   |   |-- Camera4
|   |   |-- actorsGT.mat
|   |   |-- calibration_shelf.json
|   |   |-- pesudo_gt
|   |   |   |-- voxelpose_pesudo_gt_shelf.pickle
|   |-- CampusSeq1
|   |   |-- Camera0
|   |   |-- Camera1
|   |   |-- Camera2
|   |   |-- actorsGT.mat
|   |   |-- calibration_campus.json
|   |   |-- pesudo_gt
|   |   |   |-- voxelpose_pesudo_gt_campus.pickle
|   |   |   |-- voxelpose_pesudo_gt_campus_fix_gtmorethanpred_case.pickle
|   |-- HM36

3. Training and Evaluation

The evaluation result will be printed after every epoch, the best result can be found in the log.

3.1 CMU Panoptic dataset

We train and validate on the five selected camera views. We trained our models on 8 GPUs and batch_size=1 for each GPU, note the total iteration per epoch should be 3205, if not, please check your data.

python -m torch.distributed.launch --nproc_per_node=8 --use_env run/train_3d.py --cfg configs/panoptic/best_model_config.yaml

Pre-trained models

Datasets AP25 AP25 AP25 AP25 MPJPE pth
Panoptic 92.3 96.6 97.5 97.7 15.8 here

3.1.1 Ablation Experiments

You can find several ablation experiment configs under ./configs/panoptic/, for example, removing RayConv:

python -m torch.distributed.launch --nproc_per_node=8 --use_env run/train_3d.py --cfg configs/panoptic/ablation_remove_rayconv.yaml

3.2 Shelf/Campus datasets

As shelf/campus are very small dataset with incomplete annotation, we finetune pretrained MvP with pseudo ground truth 3D pose extracted with VoxelPose, we expect more accurate GT would help MvP achieve much higher performance.

python -m torch.distributed.launch --nproc_per_node=8 --use_env run/train_3d.py --cfg configs/shelf/mvp_shelf.yaml

Pre-trained models

Datasets Actor 1 Actor 2 Actor 2 Average pth
Shelf 99.3 95.1 97.8 97.4 here
Campus 98.2 94.1 97.4 96.6 here

3.3 Human3.6M dataset

MvP also applies to the naive single-person setting, with dataset like Human3.6, to come

python -m torch.distributed.launch --nproc_per_node=8 --use_env run/train_3d.py --cfg configs/h36m/mvp_h36m.yaml

4. Evaluation Only

To evaluate a trained model, pass the config and model pth:

python -m torch.distributed.launch --nproc_per_node=8 --use_env run/validate_3d.py --cfg xxx --model_path xxx

LICENSE

This repo is under the Apache-2.0 license. For commercial use, please contact the authors.

Owner
Sea AI Lab
Sea AI Lab
GitHub Repository
A Python package to request and process seismic waveform data from Hi-net.

HinetPy is a Python package to simplify tedious data request, download and format conversion tasks related to NIED Hi-net. NIED Hi-net | Source Code |

Dongdong Tian 65 Dec 09, 2022
switching computer? changing your setup? You need to automate the download of your current setup? This is the right tool for you :incoming_envelope:

🔮 setup_shift(SS.py) switching computer? changing your setup? You need to automate the download of your current setup? This is the right tool for you

Mohamed Elfaleh 15 Aug 26, 2022
Gives criticality score for an open source project

Open Source Project Criticality Score (Beta) This project is maintained by members of the Securing Critical Projects WG. Goals Generate a criticality

Open Source Security Foundation (OpenSSF) 1.1k Dec 23, 2022
Today I Commit (1일 1커밋) 챌린지 알림 봇

Today I Commit Challenge 1일1커밋 챌린지를 위한 알림 봇 config.py github_token = "github private access key" slack_token = "slack authorization token" channel = "

sunho 4 Nov 08, 2021
Python Cheat Sheet

Introduction Pysheeet was created with intention of collecting python code snippets for reducing coding hours and making life easier and faster. Any c

CHANG-NING TSAI 7.5k Dec 30, 2022
VacationCycleLogicBackEnd - Vacation Cycle Logic BackEnd With Python

Vacation Cycle Logic BackEnd Getting Started Existing virtualenv If your project

Mohamed Gamal 0 Jan 03, 2022
This is a far more in-depth and advanced version of "Write user interface to a file API Sample"

Fusion360-Write-UserInterface This is a far more in-depth and advanced version of "Write user interface to a file API Sample" from https://help.autode

4 Mar 18, 2022
With the initiation of the COVID vaccination drive across India for all individuals above the age of 18, I wrote a python script which alerts the user regarding open slots in the vicinity!

cowin_notifier With the initiation of the COVID vaccination drive across India for all individuals above the age of 18, I wrote a python script which

13 Aug 01, 2021
Small scripts to learn about GNOME internals

gnome-hacks This is a collection of APIs that allow programmatic manipulation of the GNOME shell. If you use GNOME (the default graphical shell in Ubu

Alex Nichol 5 Oct 22, 2021
An application to see if your Ethereum staking validator(s) are members of the current or next post-Altair sync committees.

eth_sync_committee.py Since the Altair upgrade, 512 validators are randomly chosen every 256 epochs (~27 hours) to form a sync committee. Validators i

4 Oct 27, 2022
A Desktop application for the signalum python library

Signalum Desktop A Desktop application on the Signalum Python Library/CLI Tool. The Signalum Desktop application is an attempt to develop a single too

BISOHNS 35 Feb 15, 2021
Pre-commit hook for upgrading type hints

This is a pre-commit hook configured to automatically upgrade your type hints to the new native types implemented in PEP 585.

snok 54 Nov 14, 2022
An After Effects render queue for ShotGrid Toolkit.

AEQueue An After Effects render queue for ShotGrid Toolkit. Features Render multiple comps to locations defined by templates in your Toolkit config. C

Brand New School 5 Nov 20, 2022
Solcast Integration for Home Assistant

Solcast Solar Home Assistant(https://www.home-assistant.io/) Component This custom component integrates the Solcast API into Home Assistant. Modified

Greg 45 Dec 20, 2022
This is a repository built by the community for the community.

Nutshell Machine Learning Machines can see, hear and learn. Welcome to the future 🌍 The repository was built with a tree-like structure in mind, it c

Edem Gold 82 Nov 18, 2022
KeyBrowser: A program launches a browser and a keylogger at the same time, is used to retrieve a person's personal information

KeyBrowser: A program launches a browser and a keylogger at the same time, is used to retrieve a person's personal information

3 Oct 16, 2022
Hydralit package is a wrapping and template project to combine multiple independant Streamlit applications into a multi-page application.

Hydralit The Hydralit package is a wrapping and template project to combine multiple independant (or somewhat dependant) Streamlit applications into a

Jackson Storm 108 Jan 08, 2023
Generates images with semantic content from distribution A in the style of distribution B

A2B Generates images with semantic content from distribution A in the style of d

Richard Herbert 2 Dec 27, 2021
This is a simple SV calling package for diploid assemblies.

dipdiff This is a simple SV calling package for diploid assemblies. It uses a modified version of svim-asm. The package includes its own version minim

Mikhail Kolmogorov 11 Jan 05, 2023
Code and data for learning to search in local branching

Code and data for learning to search in local branching

Defeng Liu 7 Dec 06, 2022