Awesome

Cyclic Test-Time Adaptation on Monocular Video for 3D Human Mesh Reconstruction (CycleAdapt codes)

Cyclic Test-Time Adaptation on Monocular Video for 3D Human Mesh Reconstruction,
Hyeongjin Nam, Daniel Sungho Jung, Yeonguk Oh, Kyoung Mu Lee,
International Conference on Computer Vision (ICCV), 2023

Installation

• We recommend you to use an Anaconda virtual environment. Install PyTorch >=1.8.0 and Python >= 3.7.0.
• Then, run pip install -r requirements.txt. You should slightly change torchgeometry kernel code following here.

Quick demo

• Prepare data/base_data folder following below Directory part.
• Download demo files and place them into data/Demo.
• To run CycleAdapt on a custom video, please refer here.
• Run python main/adapt.py --gpu 0 --cfg asset/yaml/demo.yml.

Directory

Refer to here.

Running CycleAdapt

In the asset/yaml/*.yml, you can change datasets and settings to use.

Run

python main/adapt.py  --gpu 0 --cfg asset/yaml/3dpw.yml

To evaluate the adapted models in your experiment folder, run

python main/test.py --gpu 0 --cfg asset/yaml/3dpw.yml --exp-dir {exp_path}

Result

Refer to the paper's main manuscript and supplementary material for diverse qualitative results.

Reference

@InProceedings{Nam_2023_ICCV_CycleAdapt,  
author = {Nam, Hyeongjin and Jung, Daniel Sungho and Oh, Yeonguk and Lee, Kyoung Mu},  
title = {Cyclic Test-Time Adaptation on Monocular Video for 3D Human Mesh Reconstruction},  
booktitle = {International Conference on Computer Vision (ICCV)},  
year = {2023}  
}