search

mks0601 / Hand4Whole_RELEASE

Official PyTorch implementation of "Accurate 3D Hand Pose Estimation for Whole-Body 3D Human Mesh Estimation", CVPRW 2022 (Oral.)
MIT License
314 stars 31 forks source link

Training face-only model #6

Open yxdydgithub opened 2 years ago

yxdydgithub commented 2 years ago

Can I get the dataset settings you used to train the face-only model? In particular, which dataset are you using for trainset_3d? In your paper, you say that the "Sterling dataset" is used to evaluate 3D face-only. Can you list its directory structure?

trainset_3d = [''] trainset_2d = ['MSCOCO', 'FFHQ'] testset = ''

Thanks

mks0601 commented 2 years ago

I couldn't find good 3D datasets to train a face-only model. Instead, I just used MSCOCO and FFHQ to train a face-only model. Please visit Pose2Pose branch of this repo (https://github.com/mks0601/Hand4Whole_RELEASE/tree/Pose2Pose) and use them. I removed Sterling dataset as it does not have comprehensive images :(

yxdydgithub commented 2 years ago

Many thanks for the prompt reply.