Does taichi-ngp renderer support training nerfs?

[dhombios]

Taichi-ngp seems to be one of the few implementations of instant-ngps that doesn´t require cuda to work. However, according to the description, it seems that it just allows rendering already trained models.

Does it support training new models or are there any plans of supporting it?

[Linyou]

Hi @dhombios,

taichi-ngp currently only supports rendering, not training. To render a new scene, you'll need to train your model using ngp_pl or other PyTorch implementations and convert it using the converter.py script in the repository.

I do plan to add training support in the future, but it will take some time to implement.

Let me know if you have any questions.

[dhombios]

Thanks for replying so quickly. It's nice to hear that there are plans for implementing model training as well

I had seen ngp-pl, but it seems to use cuda. That might be a problem for running it with amd graphic cards (I haven't tried running it with non cuda pytorch implementation)