Paper reproduction: pare

Write it at the front ：

01, For purely personal learning , Do not do any business . If there is any infringement , Contact me to delete .

02, Limited personal ability , For reference only , Welcome to discuss , Grow up together .

One . Paper information

《PARE: Part Attention Regressor for 3D Human Body Estimation》

Address of thesis ：https://arxiv.org/pdf/2104.08527.pdf

Code address ：https://github.com/mkocabas/PARE

B Station introduction video

Two . Content abstract

The research content of this paper ：3D Human detection

Author related ： The boss of German Mapu Institute

The innovation and structure of the thesis ：xdm Self study

Visual model for occlusion sensitivity ,  Optimized direct regression method is used , It is mentioned in the paper that , First, three-dimensional body parameters are regressed in an end-to-end manner , Learn the attention weight of each body part , Then in order to identify or complete the occluded part , In this paper, we use the pixels of the visible part to characterize the image , To complete the image , Achieve the recognition effect .

I won't analyze the details , To put it simply, I also read the rendering of the paper , Let's show some of you .

 

On the left is the original picture , On the right is the renderings , By comparison, we can get , Completion of the occluded part , It is still highly consistent with the actual logic .

3、 ... and . Code implementation , Reason your own demo

You need to have the basic level of alchemy members .

For example, the installation environment downloads the corresponding packages , Simply speaking .

First step

git clone https://github.com/mkocabas/PARE.git

  Download the code on your server first .

The second step

Download the trained model ,

Here I put the trained model and data on the network disk , Call me Lei Feng

 link ：https://pan.baidu.com/s/1_z-Ycqaf6PMrGSwkTQMQ0A 
 Extraction code ：ndiy 

The code distribution under the directory is as follows ：

data/
├── body_models
│   └── smpl
├── dataset_extras
├── dataset_folders
│   ├── 3doh
│   └── 3dpw
└── pare
    └── checkpoints

You can compare it with mine .

  And then execute the command

cd /PARE
python scripts/demo.py --vid_file data/sample_video.mp4 --output_folder logs/demo 

That's all right. .

perfect , Crosshair .

You can show it to your teachers or friends .

 

END

-----------------------------------------------------------------------------------------------

Statement : This article is for personal study only , Because of the author's limited ability , Relevant views and information are for reference only .

If there is infringement, please contact .

Welcome readers to ask questions and exchange .

Our goal is the sea of stars ！

（ Take your brother dddd）