Clarification and help on post #39 (help modifying preprocessor targets of hands and face)

[JS1989git]

Post is based off of post asking about how to remove facial mapping found here: #39

The 3rd post directs op to a line of code in one of the .py files and it seems to have been enough information to direct op to what they need to do for what they want. However, it does not compute for me and I'm sure many others as to what technically needs to be done/edited. I have tried with my own knowledge base/savvy and come up dry.

1. If you could please be so kind as to forgive the light hand holding, would anyone please be able to tell me how I am supposed to edit the extension files so that the preprocessor will functionally produce an "openpose hand" preprocessor effect by removing the face map detection that the thread is implying is possible?

2. How can I set this up so that I then have the same kind of easy drop down menu options in the a1111 ui with those settings in mind so that I can have the rest of the dw openpose preprocessors? This is so I can functionally have not just "dw_openpose_full" as my one and only dw preprocessor, but also:

dw_openpose (body frame only)
dw_openpose_face (body frame + face map, but no hands)
dw_openpose_faceonly (though not sure is needed)
dw_openpose_hand (body frame + hands, but no face map)

Because I am assuming that in editing the .py files correctly to where they DON'T map the face, it will still be configured to function only in that one dw preprocessor model. And I'd certainly like to be able to just replicated and rename the .py files so that I don't have to manually go in and recode them when I, say, do need the face or need any other preprocessor functionality like only openpose with no face or hands.

P.S. If simpler, I would also deeply appreciate (as I'm sure many would who come across this issue) a "download these files and put them in their appropriate folders" alternative to instruction.

[JS1989git]

Please pardon the mentions @ailingzengzzz and @yzd-v but is there any chance that this might get a response? I keep coming up dry in every place I look into or inquire.. Dx

[yzd-v]

DWPose always predicts all the 133 key points together via a single model. You can add your own py file that just draws the hands or face.

[JS1989git]

Thank you for responding. But I don't have the technical knowledge to do that, it seems. Dx

Would you be able put a .py file in the repo that does that? (so it would detect body and hands only). I'm sorry boil it down to something as annoying as "can you do it for me?" but I really have no clue what to do with that information since I'm just a user/experimenting with sd and its ui and am not in the space of its coding. I can logic my way through a lot of things but this is seeming to have a lot more steps and changes that need to be made that I'm not really equipped to implement

Going to also ask @mrhanCH if he could perhaps help too since he seemed to know exactly what to do just from the highlighted line in the init file.

Is there a reason that it was just the full model that was released instead of all variations if it's just a matter of editing the code on what a preprocessor picks up? Because, again, the utility of being able to just download that with the repo would give people so much more utility with what it detects so much better than the base controlnet preprocessors. But less than 1% of those people, I would wager, would be able to modify it themselves. :\

Again, so very sorry for sounding stupid/like I accidentally stepped into advanced calculus having barely passed algebra 1, metaphorically speaking.

[mrhanCH]

Thank you for your attention.
I'm sorry, but this issue has not been verified yet. Due to busy work, I have only observed the code structure and have not DEBUG the results.
The model mentioned by the collaborator earlier inferred 133 key points, so I believe it is possible to separate hands, feet, body, and other parts through code, which requires verification.
I will also take time to DEBUG to verify the developer's proposal. Thanks.

[JS1989git]

Thank you for your attention. I'm sorry, but this issue has not been verified yet. Due to busy work, I have only observed the code structure and have not DEBUG the results. The model mentioned by the collaborator earlier inferred 133 key points, so I believe it is possible to separate hands, feet, body, and other parts through code, which requires verification. I will also take time to DEBUG to verify the developer's proposal. Thanks.

Ah ok, I see. So the information wasn't quite as cut and dry as it seemed to me. Thanks for the clarification. I look forward to seeing your future debugging.

@yzd-v Are there any plans to release the other models? I.E. body + hands, body only, etc.