Additional composition output/export type: JSON

[Balthamos]

Hi! Loving your project, it's fantastic to see it constantly evolving!

I saw you mention elsewhere that you're not able to convert 2D JSON pose data to 3D, but what I'm wondering is if it would be possible to do the opposite:

When a pose is created and the camera has been positioned and the user clicks Generate, in addition to the skeletal .png/depth/normal/canny outputs, would it be possible to generate a 2D OpenPose JSON dictionary that represents the scene's current snapshot?

Similar to this:

{
  "people": [{
    "pose_keypoints_2d":
    [321,150,1,359,232,1,293,224,1,250,300,
    1,263,365,1,425,239,1,449,354,1,427,459,
    1,294,427,1,289,505,1,-1,-1,0,372,451,1,374,
    511,1,-1,-1,0,322,134,1,338,137,1,-1,-1,0,379,
    143,1]
    },{
    "pose_keypoints_2d":
    [173,157,1,137,231,1,72,230,1,46,319,1,118,337,
    1,202,231,1,216,322,1,199,346,1,123,414,1,99,
    501,1,-1,-1,0,190,410,1,195,507,1,-1,-1,0,159,142,
    1,177,145,1,125,149,1,-1,-1,0]
     }],
  "width": 512,
  "height": 512
}

It would be very useful as some tools can't import skeleton .pngs, but can import JSON.

Thanks!

[ZhUyU1997]

interesting! I try to support it. Can you tell me which tool to import?

[Balthamos]

The tool that I use poses with is the Draw Things app for Mac/iOS, it's a fantastic fully featured Stable Diffusion generator that runs completely offline.

It can import 2D poses in OpenPose JSON format, and the poses can then be adjusted inside the app, but poses that are only .pngs are a lot more limited as they can't be edited. The JSON format that would need to be exportable by your editor is just standard OpenPose 2D output