Segment Openpose/DWpose poses ?

[InspectaGadget]

Hi !

Do you have in mind a way to segment Openpose/DWpose skeletons (in a composition based on several Openpose/DWpose characters/skeletons so), in the way to get a manipulable SEGS from each original skeletons, before generation, instead of segmenting from the generated picture ?
Cause even if the openpose preprocessor provider with controlnet on SEGS often works well, I am thinking it would sometimes be easier to edit/refine each character separately from its original openpose and disposition (like a skeleton SEGS so) in picture.
It's possible to extract keypoints (from openpose only as far as I know), trying to isolate one of several characters, but it's become hard to get it segmented.
I didn't found any ultralytics learn on open pose, and other like person ones don't manage to detect well openpose skeletons.

I took a look to the great Openpose to region map (v2) workflow, but i didn't manage to used it cause the regional ipadapter with channel penalty were replace by another version, without that setting. And in my mind, I would like to do the tricks without using Ipadapter.

Thanks by advance for your answer and for your works ! 😃

[ltdrdata]

The ipadapter part is just an example sub-workflow that utilizes the converted region map.
And if needed, you can simply replace that node with another one.