abdallahdib

plz can u share visual results and code snippest

ur training code looks fine to me. Just initialize ur fully connected layers properly. Fr instance init the focal layer with a good values (3000). same for other semantic params....

Hi I never tested with 10000 iterations, u need to increase the shape/expression regularuzation to get better fitting

Did u try multiple view points for same subject? This could improve the accuracy of the geometry. Using single image for the reconstruction is ill posed. Having multiple views improves.

U may need to improve/add more facial landmarks to obtain an accurate fitting.

Hi, the main reason why the first three Sh coeffs are not smoothed during the optimization is to capture as much as possible of the higher frequencies of the incident...

Hi here u can download the uvparam for 1K and 2K texture resolution https://github.com/abdallahdib/NextFace/releases

u can bake everything in the diffuse albedo. to do so, u can switch to a lambertian brdf by disabeling the specular and roughness components. also remove all texture regularizers...

Hi yes what u r suggesting of resizing to the target reosluton could fix that issue

can u plz give more details on the error so that i can try to help. where is the errot getting from ?