Can you share your training video? about pytorch-everybodydancenow HOT 9 CLOSED

@prismformore I'll upload my own implementation in a matter of days...

from pytorch-everybodydancenow.

@Lotayou
Here is my train dataset.
But, I don't think it is suitable for this work.
I should increase the number of images and take more various poses.
I will remove the link in a week.

from pytorch-everybodydancenow.

Thanks for the heads up. I'll keep on rolling with my work. Let's stay in touch and continue sharing info with each other. Good luck bro.

from pytorch-everybodydancenow.

Thanks!

from pytorch-everybodydancenow.

I've finished downloading the dataset. Feel free to disable the link anytime.

BTW, do you mind telling me how long does it take to train such a model (I have one Nvidia card with 12G memory)? I can use a rough timeline to coordinate with my colleagues.

Also, there's one thing puzzling me: It seems like pix2pixHD is designed to translate the segmentation masks to real images, not the skeleton stick-figures. Are we really supposed to use the skeleton as label instead of treating it as a normal RGB image from a different modality?

Anyway, I'll try using both body masks and image skeletons as input, gonna open a new repository on that too. I'll keep you notified if anything new happens.

from pytorch-everybodydancenow.

Hey, any progress? I've successfully trained pix2pix model with my own dataset and will publish the result these days

from pytorch-everybodydancenow.

@Lotayou I think the input of pix2pixHD is RGB image in this code

from pytorch-everybodydancenow.

@prismformore My thought exactly. using RGB input works well. It's just that nyoki's dataset convert skeleton to labels that I feel necessary to clarify this.

from pytorch-everybodydancenow.

@Lotayou Would u mind showing us your generated results with a comparison of different types of input (real image and skeleton)? Which one is better?

from pytorch-everybodydancenow.