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

How to test on my own data？ #6

Closed XuanchenLi closed 8 months ago

XuanchenLi commented 8 months ago

Thank you for your impressive work!

I have my own multi-view images captured by Light Stage and have tried to use your pretrained model to generate FLAME heads, but I found that the results are completely garbled. Do I need to train the model by myself even if I want to generate heads with the same topology as yours? Or is the error due to my data not being aligned properly?

TimoBolkart commented 8 months ago

Hey, TEMPEH is trained on one type of data from one scanner system and therefore not expected to generalize to another system. To test it on your camera setup I suggest you train TEMPEH on your data and then test generalization to unseen subjects and sequences of your capture system.