search

Totoro97 / f2-nerf

Fast neural radiance field training with free camera trajectories
https://totoro97.github.io/projects/f2-nerf/
Apache License 2.0
933 stars 69 forks source link

get Nan! #30

Open Khoa-NT opened 1 year ago

Khoa-NT commented 1 year ago

Hi @Totoro97,

Thank you for your great work. I tried to train your code with my own dataset.

python scripts/run.py --config-name=wanjinyou \ dataset_name= case_name= mode=train \ +work_dir=$(pwd)

But I got Nan!. Can you help me? How to modify the configuration in wanjinyou to overcome the Nan?

Thank you

Totoro97 commented 1 year ago

Hi, how is the synthesized image quality? It is OK to see one or two Nans in the console. This printing message means we find nan numbers from backward gradients and we will skip this iteration and not optimize the parameters in this training step.

Khoa-NT commented 1 year ago

Hi, thank you for reaching out.

how is the synthesized image quality?

I trained with real photos which are photos of my campus building. This is the log train_log.txt In the end, the loss is always Nan.