HumanMultiView icon indicating copy to clipboard operation
HumanMultiView copied to clipboard
verified publisher icon williamljb

Installation guide

Open lisa676 opened this issue 5 years ago • 5 comments

Hi @williamljb thanks for such a nice and amazing work. This project is really helpful.

I'm unable to find detail about project installation, can you provide detail installation about project that where to add SMPL, pre-trained weights etc...

Thanks for your support.

lisa676 avatar Mar 13 '20 03:03 lisa676

Hi Ian786,

Thanks for your interest! For installation, please check the guide here. The prerequisites are the same. For training/test options, please see src_ortho/config.py

williamljb avatar Mar 13 '20 15:03 williamljb

Hi @williamljb thanks for your kind response. I just gave it my best try and it's working only for single image.

I have some questions, kindly help me by answering these questions

  1. What do you meant by multi-view?
    (a) Getting single/multi images as input and then view results of all corresponding images in different views? I mean image 1 (input) --> result (in different views), image 2 (input) --> result (in different views). or (b) Getting multi images from different view (may be from 2 cameras) as input and then view best possible results (as one image)? I mean images (1+2) --> results (as single image having different view).

  2. In demo.py line 156 you declared a list fcs which is getting faces.obj and show.obj file. What is the purpose of this list, is it necessary part of this repository, if yes then from where we can get faces.obj and show.obj files?

  3. If I passed only single image, I can get result in two different views along with original view. It means for single image code is working fine but when I passed more than one images having different views (captured from 2 cameras with different views) then I'm getting following error.

Screenshot from 2020-03-16 16-56-01

Thanks for your precious time.

lisa676 avatar Mar 16 '20 09:03 lisa676

Hi Ian786,

  1. It is (b). The final visualized result will be on camera 0.
  2. The faces in the list are the same as SMPL models. This is just a convenient way to reconstruct and export the body mesh since I didn't know the API for the faces in the renderer or in the SMPL model.
  3. Can you show me your command? You can check if the image paths are parsed correctly here. I guess one possible reason is that you may have an extra comma at the end of the input string.

williamljb avatar Mar 16 '20 15:03 williamljb

Hi @williamljb here is my command python -m demo --img_paths data/image1.png, data/image2.png

lisa676 avatar Mar 17 '20 01:03 lisa676

Hi Ian786,

The problem is because there is an extra space between two image paths. Try the following: python -m demo --img_paths data/image1.png,data/image2.png

williamljb avatar Mar 17 '20 15:03 williamljb