search

TimoBolkart / TF_FLAME

Tensorflow framework for the FLAME 3D head model. The code demonstrates how to sample 3D heads from the model, fit the model to 2D or 3D keypoints, and how to generate textured head meshes from Images.
http://flame.is.tue.mpg.de/
444 stars 77 forks source link

About build_texture_from_image.py #26

Closed Ten1o closed 4 years ago

Ten1o commented 4 years ago

Hi, Thanks for your excellent work!

I've been trying to figure out how to use an image to build texture to the FLAME mesh (using build_texture_from_image.py) and I had several questions:

  1. I saw the file needs to take in a pre-computed parameter "target_scale". I'm wondering how to calculate this parameter if I want to align my own image with its FLAME mesh.

  2. I noticed that the example meshes given (such as imgHQ00039.obj) are not zero-centered. They have offsets from the origin and it seems that this offset is useful for calculating the texture map. How is this offset chosen?

Thanks for answering!

TimoBolkart commented 4 years ago

Hi,

The association of FLAME with those images is established in the fit_2D_landmarks.py. Both, the model translation and the scale are parameters that are optimized within the fit_lmk2d function by minimizing the reprojection loss between the (projected) model landmarks and the predicted image landmarks. Does this answer your question?

Ten1o commented 4 years ago

Yes! Thanks for your answer. I will check that file!

zhangqijun commented 4 years ago

I'm sorry ,where is imgHQ00088.obj

Ten1o commented 4 years ago

I'm sorry ,where is imgHQ00088.obj

I'm not sure but you can check https://github.com/TimoBolkart/TF_FLAME#fit-2d-landmarks , which may be used to generate the obj file.