TimoBolkart

Yes, that makes perfectly sense. We commonly follow a similar optimization strategy.

Hello, FLAME uses linear blendskinning (LBS) to articulate the head rotation around the neck, the jaw rotation, and to rotate left and right eye. The model.weights are the LBS blendweights...

You are right, estimating eye gaze by optimizing FLAME's eyeball pose parameters requires additional landmarks for the pupil center. If you are using a landmark predictor that provides pupil center...

Fitting the model to a scan minimizes follows some iterative closest point scheme, which alternates between computing correspondences between each scan vertex and the FLAME model mean, and minimizing the...

You are right, the placement of the 51 landmarks is designed to capture expressions not shape. FLAME can capture a large variation of face shapes, but not from landmarks alone....

Hello, can you please specify what you are running? The file path hints at something related to Imitator, and voca_face_former is nothing that is distributed with the VOCA repository. Did...

You can download all data from the VOCA website (voca.is.tue.mpg.de). For access, you must register and accept the license.

You are right, first, the Chumpy code is not super efficient, and second, 1M vertices is huge. I suggest you either downsample the mesh to a lower resolution, or to...

The data are 3D performance captures from our multi-view capture system. The participants can freely move their head during the captures which result in variation in head position and orientation....

Hi, as all meshes are in dense topological correspondence to FLAME, computing the parameters can directly be computed by minimizing some vertex-to-vertex loss between FLAME and each registered mesh.