rover-xingyu / l2g-nerf Goto Github PK

Thanks for your great job! During reading your code, I have a problem can't figure out.

Why don't you use camera_cords_grid_3D directly, but transfer to camera_cords_grid_2D before put in warp_mlp？ Is that necessary?

Thanks again! :)

Thanks for releasing the code, I really appreciate your work!

I tried to run the code with l2g_nerf model, and the value of taylor series shoots to +/-infinity in camera.se3_to_SE3 .

It seems that this is caused by the warp_mlp giving very large (about 100~1000) se3 values without initialization.

How should I initialize warp_mlp and warp_embedding? Or did I miss anything?

Thanks :)

Have you encountered any peculiar phenomena when using Umeyama's method? For instance, in certain situations, have you observed significantly large distances between the estimated poses aligned using Umeyama's method and the ground truth poses?

Traceback (most recent call last):
File "train.py", line 35, in
main()
File "train.py", line 32, in main
m.train(opt)
File "G:\work_document\python_work\L2G-NeRF-main\model\nerf.py", line 61, in train
if self.it%opt.freq.val==0: self.validate(opt,self.it)
File "G:\work_document\Tensorflow\Miniconda3\envs\L2G-NeRF\lib\site-
packages\torch\autograd\grad_mode.py", line 27, in decorate_context
return func(*args, **kwargs)
File "G:\work_document\python_work\L2G-NeRF-main\model\l2g_nerf.py", line 89, in validate
super().validate(opt,ep=ep)
File "G:\work_document\Tensorflow\Miniconda3\envs\L2G-NeRF\lib\site-packages\torch\autograd\grad_mode.py", line 27, in decorate_context
return func(*args, **kwargs)
File "G:\work_document\python_work\L2G-NeRF-main\model\base.py", line 154, in validate
loss = self.summarize_loss(opt,var,loss)
File "G:\work_document\python_work\L2G-NeRF-main\model\base.py", line 139, in summarize_loss
assert not torch.isnan(loss[key]),"loss {} is NaN".format(key)
AssertionError: loss render is NaN

training: 0%| | 0/200000 [00:00<?, ?it/s][val] loss:2.060e-01
training: 1%| | 1998/200000 [01:59<3:14:20, 16.98it/s, it=2000, loss=0.020]
training: 1%| | 1998/200000 [02:18<3:14:20, 16.98it/s, it=2000, loss=0.020]

The progress bar didn't move for three hours!

Hi, thanks for this amazing work and publishing the code alongside it!
Right now, you don't specify any license for the code in this repository (as far as I can see) - is this something that you could add so that people know how they're allowed to use the code?

All the best,
Stefan

Hi, I noticed the experimental results of BARF in the paper are very different from that in the original BARF paper. I wander if it is because the parameter setting ?

您好，感谢您的工作。
我在复现的时候遇到了一些问题，有关warp网络的初始化，和它的激活函数。想问一下是怎么设置的？
再次感谢！

Hi! Thanks for your awesome work ! I want to know when the code will be released . I can not wait to test your wodenful work!!!

Is there a way to get vertex colors while extracting the mesh?

What do you mean by Translation errors are scaled by 100.?
I want to know the unit of translation errors?
Is it meter?