Would be nice to have a requirements.txt file to ensure no dependencies or critical packages are missed.

[harshakmohan]

No description provided.

[tares003]

what are the requirements?

[jeremyfix]

From what I can tell for now, the code is running if you use the following requirements :

torch==2.2.1
tqdm
imageio>=2.34.0
numpy>=1.24.4
opencv-python-headless>=4.9.0.80
pyvista>=0.43.3
kornia>=0.7.1
configargparse

I'm able to execute the code on the chair example : python run_nerf.py --config configs/chair.txt --finest_res 512 --log2_hashmap_size 19 --lrate 0.01 --lrate_decay 10

I nevertheless had to bring some modifications .

• in run_nerf.py, replace set_default_tensor_type by torch.set_default_device(device) ; Not sure If we have to enforce the dtype as well
• in run_nerf.py, replace torch.Tensor([1e10]) (which was on CPU for some reasons, despite the set_default_device ?!) by torch.tensor([1e10]) (which correctly used the default device)
• do the same as above in run_nerf_helpers.py for the ms2psnr lambda function

edit: I believe the code is running , the PSNR is increasing; and here is an example frame produced at iteration 2000

and here at 4000 iterations

[rib12316]

@jeremyfix,Thank you for your answer, I was having the same problem of two variables not being able to be calculated on cpu and cuda respectively, I would like to add that it is possible to set a breakpoint on pychram to check which variable is on the cpu, I just checked that c2w on line 40 in the file utils.py was on the cpu and just replace
c2w = torch. FloatTensor(frame[‘transform_matrix’]) by
c2w = torch.FloatTensor(frame[‘transform_matrix’]).to(‘cuda:0’)
and finally it works！