PRNet icon indicating copy to clipboard operation
PRNet copied to clipboard

There is a problem with the face depth map

Open FUWhy opened this issue 2 years ago • 2 comments

I tried to generate some depth maps, but the results were strange, the depth maps seemed to be the same across different faces. My anconda environment is as follows: Ubuntu20.04 ,docker based python 3.6.13 tensorflow-gpu 1.15.0 dlib 19.24.99 opencv-python 4.7.0.72 I tested it directly using demo.py,The results are as follows: test 1: 微信图片_20230307170910 微信图片_20230307170923 test 2: 微信图片_20230307170928 微信图片_20230307170932 test 3: 微信图片_20230307170937 微信图片_20230307170942 I tried to solve the problem, but without success. I don't know what the problem is,I hope you can get some help. Thank you.

FUWhy avatar Mar 07 '23 09:03 FUWhy

I ran into similar issue, when using --isDlib False. When I keep it True, I get various depth maps.

Veronika-Polakova avatar Aug 11 '23 08:08 Veronika-Polakova

I had the same problem and solved it by downgrading TensorFlow "1.14" to TensorFlow "1.4.1" with CUDA 8.0. I don't really know what the problem was, but now it works as expected.

biesseck avatar Feb 13 '24 13:02 biesseck