search

ltdrdata / ComfyUI-Impact-Pack

Custom nodes pack for ComfyUI This custom node helps to conveniently enhance images through Detector, Detailer, Upscaler, Pipe, and more.
GNU General Public License v3.0
1.91k stars 188 forks source link

Segment Openpose/DWpose poses ? #786

Open InspectaGadget opened 1 month ago

InspectaGadget commented 1 month ago

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 commented 1 month ago

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.