Facebook Reality Labs Pittsburgh is working on realistic 3D face reconstruction at VR headsets. They built a dome to collect multi-view human face data and developed an accurate geometry-based 3D reconstruction algorithm. However, they found two problems with their pipeline. First, they don’t have ground truth data to evaluate it. Second, they need the algorithm to be faster. Our project is set to tackle these two problems. We built a synthetic dataset to evaluate the existing pipeline and explored several deep learning methods to achieve acceleration.
You can find a summary of our work as well as demos in our final presentation slides:
And here are our PPF progress reports:
Contacts:
He Wen (hew1@andrew.cmu.edu)
Ruoyuan Zhao (ruoyuanz@andrew.cmu.edu)