Ruiqi LI comments

Results 9 comments of


                                            Ruiqi LI

Can't download the dataset since 201

Same issue. 201-300.tar.gz and previous files work, but the subsequent files don't.

jt.mean produces computation results with insufficient floating point precision

Same result in my case, which leads to imprecise SSIM.

train_ngp_nerf_occ.py: RuntimeError: CUDA error: invalid configuration argument

Same problem in version 0.5.2, volrend.accumulate_along_rays can not handle empty input

Code for initializing NeRF's geometry as a sphere

Refer to the geometric network initialization method from this paper: SAL: Sign Agnostic Learning of Shapes from Raw Data

Can not open http://0.0.0.0:7860

Try setting block.launch(server_name='localhost')

OOM during backward

Thanks for the prompt reply! I tried setting dynamic_ray_sampling to false and train_num_rays=128 however OOM still happened. Yes I did not implement the background model, but I utilized the foreground...

OOM during backward

> Could you share the training data so that I can debug? Thanks a lot! I've emailed your :)

OOM during backward

> The problem is gone. Many thanks!

code release

Still looking forward to the code!