search

quan-meng / gnerf

[ ICCV 2021 Oral ] Our method can estimate camera poses and neural radiance fields jointly when the cameras are initialized at random poses in complex scenarios (outside-in scenes, even with less texture or intense noise )
MIT License
227 stars 27 forks source link

GIF result on hotdog & lego dataset: all is white #7

Open ZhangXiaoXuan2019 opened 1 year ago

ZhangXiaoXuan2019 commented 1 year ago

Hello authors, I have read the GNeRF paper recently and try to re-product the results on blender hotdog and lego dataset, with released code in this repository and default settings. But I find that the output gif result is all white after at least 30k training iterations.

I have also read the previous issues proposed previously here. The author say that it happens that the GAN part training fails and leads to all-white results. I wonder whether it is normal for the GNeRF to fail on GAN training?

quan-meng commented 1 year ago

Yes, the gan training sometimes will fail because the density field becomes all empty and generate all white images, it is related to some parameters like image size, dataset type, etc. I still cannot guarantee that it will be stable in all the scenes with same parameters :)

I often look at the first hundred iterations and restart the training if an imbalance occurs between the generator and discriminator.