search

sxyu / plenoctree

PlenOctrees: NeRF-SH Training & Conversion
https://alexyu.net/plenoctrees
BSD 2-Clause "Simplified" License
423 stars 78 forks source link

own synthetic data, loss is not converging. #7

Open g-gaurav opened 3 years ago

g-gaurav commented 3 years ago

Thanks for the great work!

I want to train the nerf-sh model on my own synthetic dataset. I have tried so many times with different dataset generated through blender script. but every time loss is not converging and gives blank white images only. Please suggest.

huying163 commented 2 years ago

请问您是怎样合成数据集的？问题解决了吗

huying163 commented 2 years ago

How to train on your own synthetic dataset? Did you sovle the probelm

saunair commented 2 years ago

Make sure your camera pose conventions match with the assumptions. I was able to get a pyrender-ed data to work. The camera poses have the Z-axis pointing outwards, and Y-axis pointing upwards (a 180 degree rotation over x-axis). I'm not sure why most of the NeRF code bases follow this different convention.