HiddenPose

Repository of paper HiddenPose: Non-Line-of-Sight 3D Human Pose Estimation.

Code will release after the paper is published.

Email: liuping@shanghaitech.edu.cn

This code repository is for paper HiddenPose: Non-Line-of-Sight 3D Human Pose Estimation （paper pdf）

Prerequisites

We run our code in Ubuntu 18.04.6 LTS with Pytorch 1.7.1

Here is one example of building development environment with conda:

conda create -n hidden_pose python=3.8
conda activate hidden_pose
conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=11.0 -c pytorch
python -m pip install tensorboard
python -m pip install yacs
python -m pip install scipy
python -m pip install einops
python -m pip install opencv-python
python -m pip install matplotlib
python -m pip install timm
python -m pip install torchsummary
python -m pip install wandb

Data

Download link is HiddenPose - Google drive

HiddenPose dataset including two parts.

1. simulation data

Google drive: HiddenPose/pose_v2_noise

The data should be origanalzed as :

pose_v2_noise
├── pose_00
│   ├── test
│   │   ├── joints
│   │   ├── meas
│   │   └── vol
│   ├── train
│   │   ├── joints
│   │   ├── meas
│   │   └── vol
│   └── val
│       ├── joints
│       ├── meas
│       └── vol
├── pose_01
│   ├── test
│   │   ├── joints
│   │   ├── meas
│   │   └── vol
│   ├── train
│   │   ├── joints
│   │   ├── meas
│   │   └── vol
│   └── val
│       ├── joints
│       ├── meas
│       └── vol
├──......
└── pose_08
    ├── test
    │   ├── joints
    │   ├── meas
    │   └── vol
    ├── train
    │   ├── joints
    │   ├── meas
    │   └── vol
    └── val
    │    ├── joints
    │    ├── meas
    │   └── vol  
    └── val
        ├── joints
        ├── meas
        └── vol

2. measured data

Google drive: HiddenPose/measured_data

Running the codes

Train

python ./train.py

Model Weights

Download link is HiddenPose - Google drive

Test

• Real data from HiddenPose

python test.py --test test_realdata --model ./trained_weights/NlosPose_final_dict_14.pth --device 0

• Test data from HiddenPose

python test.py --test test_pose_v2 --model ./trained_weights/NlosPose_final_dict_14.pth --device 0