NameError: name 'tcnn' is not defined

[Amaru-Zeas]

I was able to train my nfactor just fine, then when I tried to use nerf2nerf it give me an error

ns-train in2n --data L:\Photogametry_Project\NeRF\amaru --load-dir C:\Users\Cuy\nerfstudio\outputs\amaru\nerfacto\2023-04-25_092552\nerfstudio_models --pipeline.prompt {"prompt"} --pipeline.guidance-scale 7.5 --pipeline.image-guidance-scale 1.5

NameError: name 'tcnn' is not defined

[tancik]

Can you try running pip install ninja git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch

[Amaru-Zeas]

thanks I I will give it a try

[Amaru-Zeas]

Thanks @tancik it seems that worked, just need to figure it out how to use the prompt in nerfstudio now :)

[Amaru-Zeas]

[Amaru-Zeas]

so training model keeps resetting back to 0% then goes all the way to 100% is this normal?

[ayaanzhaque]

What do you mean resetting to 0? If you load the nerfacto model with in2n, it will start the training at 100% (just because the pre-trained model has already been trained for 30k iters). This is expected.

[Amaru-Zeas]

[Amaru-Zeas]

see how from 97% goes back to 00%

[Amaru-Zeas]

I guess I am still confused how to use in2n and the prompt in nerfstudio

[ayaanzhaque]

Oh lol. Its going from 103.97% to 104.00%. The training is going properly. You can set the prompt using the command line, as described in the readme.

ns-train in2n --data {PROCESSED_DATA_DIR} --load-dir {outputs/.../nerfstudio_models} --pipeline.prompt {"prompt"} --pipeline.guidance-scale 7.5 --pipeline.image-guidance-scale 1.5

[Amaru-Zeas]

lol gotcha didn't know goes beyond 100%

[Amaru-Zeas]

[Amaru-Zeas]

so I do have the prompt in nerfstudio, so just add anything in that box and wait for it?

[ayaanzhaque]

The prompt that you added via the command line will show up in that text box. If you want to adjust the prompt during training you can do so. Otherwise, all you have to do is run the command line and then let it train.

[Amaru-Zeas]

ah I see thanks, just taking long time to train with in2n (the regular train was much faster) I am using a RTX 4090. so if currently is 107.13% what percentage is completed?

[ayaanzhaque]

Ya it takes a lot longer to train than the regular nerfacto because it has to use a diffusion model. It usually will be done around 40k iterations. I would say just check on it in the viewer from time to time to see how the edit is progressing.

[Amaru-Zeas]

ah I see thanks a lot. One last question, so everytime I add a different prompt while training will that affect time?

[ayaanzhaque]

Ya it will make it take a while to converge to your new edit.