search

fudan-generative-vision / champ

Champ: Controllable and Consistent Human Image Animation with 3D Parametric Guidance
https://fudan-generative-vision.github.io/champ/
MIT License
3.45k stars 408 forks source link

reference image #132

Open yankainan opened 1 month ago

yankainan commented 1 month ago

Thank you for your excellent work! I would like to ask, what is your strategy for selecting a reference image from a video clip? For example, do you select the first frame of a video clip?

AricGamma commented 3 weeks ago

Which step are you dealing with?

wanghaisheng commented 3 weeks ago

`-- ref_images/ # Reference image samples(Optional) why @AricGamma this is optional, can you explain a little how to get and metric of this reference image

yankainan commented 2 weeks ago

Which step are you dealing with?

that isn't in data_process.md, just a strategy of selecting reference images

AricGamma commented 1 week ago

Which step are you dealing with?

that isn't in data_process.md, just a strategy of selecting reference images

Or what do you want to do with the selected reference image? Training? Inference? Or sth. else?

AricGamma commented 1 week ago

`-- ref_images/ # Reference image samples(Optional) why @AricGamma this is optional, can you explain a little how to get and metric of this reference image

ref_images folder is optional logically. But we coupled video SMPL and ref images preprocessing together. You can ignore the Optional annotation for now. We will optimize the data processing logic.

wanghaisheng commented 1 week ago

@AricGamma how to gen such ref_images does this have some minimal requirements