search

xiaoxingzeng / DF2Net

DF2Net: A Dense-Fine-Finer Network for Detailed 3D Face Reconstruction
122 stars 17 forks source link

About DNet #5

Closed ava-YangL closed 4 years ago

ava-YangL commented 4 years ago

Hi Xiaoxing, When I use the DNet code you released to train my model, the 3D face I get is not smooth, as shown in the figure below. image

The depth value of my 3D face is about 0 to 300, and the image size is 512 * 512. I wonder if you have regularized the depth value during training. Could you please tell me the range of your depth value or give me some suggestions. Looking forward to hearing from you. Thanks.

xiaoxingzeng commented 4 years ago

Hi Yang , thanks for your interest. The range of depth value is [0, 1], actually we got the image-depth pairs with the code from https://talhassner.github.io/home/publication/2014_MVAP. In additional, You can add a smooth term.