Is Scene box working correctly?

[snarb]

I have noticed that changing

scene_box = SceneBox(aabb=torch.tensor([[-1.5, -1.5, -1.5], [1.5, 1.5, 1.5]], dtype=torch.float32))

in the blender_dataparser.py to

scene_box = SceneBox(aabb=torch.tensor([[-.005, -.005, -.005], [.005, .005, .005]], dtype=torch.float32))

does not affects the training on the Lego scene significantly. This looks suspicious to me. Maybe this just means that normalization from 0 to 1 doesn't meters a lot. (Collider, camera optimizer, and average appearance embeddings are disabled.Nerfacto)

[tancik]

Since there is a spatial_distortion in the nerfacto model, the scene box is never used. https://github.com/nerfstudio-project/nerfstudio/blob/009e21be248418028e6303ecde593a20ef88fa87/nerfstudio/fields/nerfacto_field.py#L199-L204 It is worth noting that the nerfacto model isn't really setup to work with the blender scenes.

[snarb]

Clear. Thanks . Does it make sense to tune nerfacto for synthetic scenes or instant ngp will work thr same or better after all?