Is Block-NeRF amenable to instant-ngp optimizations? about unboundednerfpytorch

[sjtuytc]

Thanks for your suggestions! I will definitely look into it and catch up with these awesome techniques!

[sjtuytc]

Hi, I'm currently studying instant-ngp as well because currently, we have to rely on some tools from instant-ngp to create interactive scenes. As for development, incorporating instant-ngp is not very easy because it is dependent on CUDA and requires building every time. Besides, it needs a desktop version of GPU but we typically only have headless servers.

[corbt]

Besides, it needs a desktop version of GPU but we typically only have headless servers.

FYI, I've successfully run renders headlessly as described in the FAQ.

it is dependent on CUDA and requires building every time

This is fair. There's a pytorch port I've got running that doesn't require any compilation but the resulting NeRFs end up with more artifacts than when I run the same collections through instant-ngp. It certainly converges fast though! 😃

[ElliotQi]

How is it going? I'm also interested in the Block-Nerf with instant-ngp.