CMU-Perceptual-Computing-Lab / MonocularTotalCapture

Code for CVPR19 paper "Monocular Total Capture: Posing Face, Body and Hands in the Wild"
http://domedb.perception.cs.cmu.edu/mtc.html
662 stars 123 forks source link

Run without visualization? #20

Closed ZhengyiLuo closed 4 years ago

ZhengyiLuo commented 5 years ago

Would it be possible to run with visualization and directly write the result to disk? I am running this in a docker image and getting the OpenGL to work proves to be troublesome.

Thanks a ton!

xiangdonglai commented 5 years ago

There are 2 possibilities as far as I know.

  1. In 'run_fitting.cpp', you can comment out all lines that contain 'Renderer' or 'render', or the lines that read outputs from the renderer. In this way you will still get output in terms of Adam parameters in 'txt' files, but not the visualized images.

  2. The other way is to substitute the OpenGL renderer with EGL, but you will need to do some engineering yourself there.

I think solution 1 is in general easier. because once you save the Adam model parameters and reconstruct the result to a mesh as I did here, you can visualize the results using any other graphical tools that can render a mesh. Hope this will help you.

wtnan2003 commented 4 years ago

@ZhengyiLuo would it be possible for you to share a dockerfile or docker image with me? thx!

nopeiyu commented 4 years ago

There are 2 possibilities as far as I know.

  1. In 'run_fitting.cpp', you can comment out all lines that contain 'Renderer' or 'render', or the lines that read outputs from the renderer. In this way you will still get output in terms of Adam parameters in 'txt' files, but not the visualized images.
  2. The other way is to substitute the OpenGL renderer with EGL, but you will need to do some engineering yourself there.

I think solution 1 is in general easier. because once you save the Adam model parameters and reconstruct the result to a mesh as I did here, you can visualize the results using any other graphical tools that can render a mesh. Hope this will help you.

I modified the code according to your method 1. It can run successfully on the remote server, but the output result is very poor. It seems that render is necessary for optical flow. How can I modify it to run correctly on the remote server? I look forward to your reply