search

mks0601 / I2L-MeshNet_RELEASE

Official PyTorch implementation of "I2L-MeshNet: Image-to-Lixel Prediction Network for Accurate 3D Human Pose and Mesh Estimation from a Single RGB Image", ECCV 2020
MIT License
708 stars 130 forks source link

Camera coordinates #95

Open bavasilev opened 2 years ago

bavasilev commented 2 years ago

Hello! The distance from the mesh is estimated by RootNet, but how to get the rotation of the camera used in the visualization of the mesh? Looking forward to getting any kind of advice.

mks0601 commented 2 years ago

The global rotation is estimated by I2L-MeshNet.

bavasilev commented 2 years ago

Could you please pinpoint the place in the code where I can get it from? I was trying to figure it out on my own but failed to do so, as I'm not sure what format it is presented in.

mks0601 commented 2 years ago

In the lixel stage, I2L-MeshNet predicts in (x,y pixel, z depth) space mesh vertices, so no global rotation is required. In the param stage, see here: https://github.com/mks0601/I2L-MeshNet_RELEASE/blob/4c71b89713eccfb893607963696f368f4beb9fa6/common/nets/module.py#L128

bavasilev commented 2 years ago

Thanks a lot!