Skip to content

Latest commit

 

History

History
89 lines (68 loc) · 4.09 KB

README.md

File metadata and controls

89 lines (68 loc) · 4.09 KB
Raw

Metrical Photometric Tracker

Wojciech Zielonka, Timo Bolkart, Justus Thies

Max Planck Institute for Intelligent Systems, Tübingen, Germany

Video  Paper  Project Website  Email


From the left: RGB Input, Estimated Texture, Predicted Landmarks (red), FLAME Geometry.
Official Repository for ECCV 2022 Metrical Photometric Tracker (from MICA).

Description

Metrical monocular tracker based on sequential color optimization. It returns optimized FLAME mesh, statistical texture coefficients, and pinhole camera intrinsic and extrinsic parameters. For each frame, a checkpoint will be created that stores all the information. Additionally, depth maps and meshes are also saved.

Installation

After cloning the repository please install the environment by running the install.sh script. It will prepare the tracker for usage. Please note to obtain BFM texture, which was used in our results and projects, you have to follow the BFM_to_FLAME repository. By default, the FLAME texture will be downloaded and used in the output folder.

Before installation, you need to create an account on the FLAME website and prepare your login and password beforehand. You will be asked to provide them in the installation script.

git clone https://github.com/Zielon/metrical-tracker.git
cd metrical-tracker
./install.sh

Usage

Our tracker needs MICA predictions to run. The identity.npy file you can find in the output folder of the demo.py file. Once the shape/identity file is generated you can simply select the corresponding video and run the tracker. Look at the three sequences {duda, justin, wojtek} from the example dataset in the input folder. Please, follow the same naming convention for your custom datasets. In the configuration file, you can specify the input and output folders.

python tracker.py --cfg ./configs/actors/duda.yml

Tips

  • in the case of fast-changing sequences you might adjust the learning rate for R and T in the config.
  • MediaPipe sometimes detects wrong landmarks for the first frame in the sequence. You might skip this frame in optimization.
  • there are many hyper-parameters involved in the optimization, please look at the config.py and tweak them to further improve your results.
  • the keyframes are the frames for which global color and shape will be optimized, therefore select frames with neutral pose preferably.

Projects

This tracker has been used in the following projects:

Citation

If you use this project in your research please cite MICA:

@proceedings{MICA:ECCV2022,
  author = {Zielonka, Wojciech and Bolkart, Timo and Thies, Justus},
  title = {Towards Metrical Reconstruction of Human Faces},
  journal = {European Conference on Computer Vision},
  year = {2022}
}