Open neekolas opened 1 year ago
Note that nerfstudio is implemented in PyTorch rather than JAX, and it can be a lot of work to migrate.
NerfStudio looks pretty modular to me. You can have a model that gets trained in a totally separate process, and as long as you can render the outputs of that model from Nerfstudio you can take advantage of the renderer and viewer (which is probably what 90% of people care about). Not saying that makes it trivial, but it does lower the scope of work quite a bit.
Nerfstudio allows for external implementations of models. It would be great if there was a version of these models compatible with nerf-studio, similar to what Tetra-Nerf has done.