search

TimoBolkart / TEMPEH

TEMPEH reconstructs 3D heads in dense semantic correspondence from calibrated multi-view images in about 0.3 seconds.
https://tempeh.is.tue.mpg.de
Other
116 stars 7 forks source link

Talking Head #1

Closed pranavmalikk closed 1 year ago

pranavmalikk commented 1 year ago

Absolutely great work and one of my favorite papers this year! Is there any plan to further expand the dataset or make a new dataset to include dialogue with matching 3D reconstruction? Or possibly utilizing some obstruction to match in the wild dataset?

TimoBolkart commented 1 year ago

Hello, thank you. We have the VOCA dataset, which is a paired dataset of speech and 3D head motion. Other than this, we are not planning to capture a large scale dialogue dataset in our 4D capture system. However, we are planning to release some facial performances for in-the-wild speech sequences later this year, stay tuned.