search

facebookresearch / StyleNeRF

This is the open source implementation of the ICLR2022 paper "StyleNeRF: A Style-based 3D-Aware Generator for High-resolution Image Synthesis"
955 stars 91 forks source link

Inference on real images. #14

Open muhammad-ahmed-ghani opened 2 years ago

muhammad-ahmed-ghani commented 2 years ago

Hi @MultiPath , how to give initial image as your own real image and then perform editing on it.

MultiPath commented 2 years ago

Hi, I haven't cleaned up the code for inversion yet. I have some scripts used in our paper previously but it is not super clean. https://github.com/facebookresearch/StyleNeRF/blob/main/apps/inversion.py Based on my experience, it is better to train an encoder first to estimate the camera and then perform latent optimization via back-propagation. Also, I found it is relatively more difficult to invert compared to the original stylegan which needs more research.

Michaelwhite34 commented 2 years ago

Any update about this ?

li-fang commented 1 year ago

Hi, have anyone done it yet?

Jain-Archit commented 1 year ago

I have a set of different view of an object and want to create a video like a 360degree nerf. Can anyone help me how to go about this?