Results on NeRF 360 Dataset

[ch-ho00]

Hi. Thank you for open-sourcing the great work!
I want to know how the results on the project pages are implemented.
Because based on the code, LLFF and NeRF 360 dataset uses one view for test (closest to the average pose) and the rest for training.
I was wondering if that is the case for reported DCVGO results.

Thank you in advance

[sjtuytc]

What do you mean by one view for the test? I think the dataset split of DVGO follows previous works.

[ch-ho00]

My understanding is that mip-nerf360 dataset is set to be a dataset type LLFF based on the below line.
https://github.com/sunset1995/DirectVoxGO/blob/main/configs/nerf_unbounded/nerf_unbounded_default.py#L6

And LLFF datasets only leave on view out for validation/test.
https://github.com/sunset1995/DirectVoxGO/blob/main/lib/load_llff.py#L416

[sjtuytc]

I'll see this later. But since dvgo is already a good baseline, We can just follow it.

[ch-ho00]

I'm not saying that DVGO is not a good baseline. I'm just making sure the reported result is based on only one view because the dataset contains a 360 degree scene so testing only on one view seems odd for me.