search

YuelangX / LatentAvatar

A PyTorch implementation of "LatentAvatar: Learning Latent Expression Code for Expressive Neural Head Avatar"
MIT License
103 stars 8 forks source link

Pose and Scale in dataset #4

Open SuperStacie opened 1 year ago

SuperStacie commented 1 year ago

Hi, thanks for the interesting work!

I wonder if the pose and scale in the dataset are being used during the training and inference. And how do we prepare the data acquire the camera parameters if we wanna try it out using self-captured data?

Thanks

oijoijcoiejoijce commented 1 year ago

+1

iperov commented 1 year ago

Yeah, they didn't specify in the paper that the dataset requires the calculation of pose and some sort of scale for each frame that go into training in Nerf. It looks like these values are calculated somehow from landmarks. Given that their examples have shaky frames, it looks like pose and scale play a key role ?

iperov commented 1 year ago

but according https://github.com/YuelangX/LatentAvatar/blob/main/lib/module/HeadModule.py they use pose+scale only color_mlp in nerf, but not in density_mlp. Perhaps we don't need that embedding?

wuzuyin commented 11 months ago

Hi, thanks for the interesting work!

I wonder if the pose and scale in the dataset are being used during the training and inference. And how do we prepare the data acquire the camera parameters if we wanna try it out using self-captured data?

Thanks

Hello, have you resolved this issue？

wuzuyin commented 11 months ago

+1 Hello, have you resolved this issue？