Giters

xk-huang / nerf.mindspore

Neural radiance field with mindspore. w/ checkpoint, re-imp. performance, and ipynb to play around.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Can you give me the correct training command?

jianglh-WHU opened this issue · comments

commented

Hi～Thanks for your excellent implemention.

But when I actually run it, I find that the code fails to converge and runs much slower than pytorch's version (about 60h in 1*A100). And my version of mindspore is 2.1.0.

I'm wondering if it's a problem with the version or the environment, or if it's a problem with the command I'm running.

python train.py --name chair_coarse --data_dir data/nerf_synthetic/materials/ --dataset_type blender --half_res --cap_n_rand 4096 --cap_n_samples 192 --cap_n_importance 0 --use_view_dirs --raw_noise_std 1e0 --white_bkgd --gpu 1

So can you me some suggestions? Thanks again!

commented

Hi @jianglh-WHU,

Please install the corresponding mindspore according to https://github.com/xk-huang/nerf.mindspore/blob/main/environment.yaml.