facebookresearch / StyleNeRF

This is the open source implementation of the ICLR2022 paper "StyleNeRF: A Style-based 3D-Aware Generator for High-resolution Image Synthesis"
956 stars 92 forks source link

Questions about the Camera Pose #25

Open FeiiYin opened 2 years ago

FeiiYin commented 2 years ago

Thanks for your impressive work. During the inversion, I wondered why we need re-train a network to predict the camera pose (pitch & yaw). Since there exist works focusing on predicting the head pose, can the predicted value from the other pre-trained models be used directly on the StyleNeRF?

However, in my experiment, I found that the predicted pose values range (DECA 3dmm) greatly differ from the camera pose used in StyleNeRF, which leads to poor inversion performance.

Thanks in advance!