Heathcliff-saku/ViewFool_

Argparse Problem

Opened this issue · 2 comments

Thanks for your great work! When trying to reproduce your work, I cannot train your vanilla NeRF due to lacking of opt.py which exists in nerf_pl; I tried to use datasets/opts.py instead but the params are not pretty matching. Directly using opt.py from nerf_pl also doesn't work. Is there some mismatch between the nerf_pl project and yours about argument parsing? If so, how should I modify the argparse code? Thank you in advance!

Hi Ep11phany,Thanks for your feedback!
Actually the datasets/opts .py are used for adversarial viewpoint optimization. Sorry that due to my negligence, I forgot to upload the opt.py for training NeRF, but directly using the opt.py in the nerf_pl should work normally, maybe you should modify the --root_dir and other parameters inside? I'll upload the opt.py file later and you can also directly use the nerf_pl to generate NeRF .ckpts files

With using opt.py, it seems like hparams doesn't have N_emb_xyz and some other essential member params anymore. How should I fix it?