Calculate bounding box?

[AIBluefisher]

Hi, @liruilong940607

I noticed nerfacc has a piece of code for calculating a rough bounding box:

    if args.auto_aabb:
        camera_locs = torch.cat(
            [train_dataset.camtoworlds, test_dataset.camtoworlds]
        )[:, :3, -1]
        args.aabb = torch.cat(
            [camera_locs.min(dim=0).values, camera_locs.max(dim=0).values]
        ).tolist()
        print("Using auto aabb", args.aabb)

The bounding box calculated this way can potentially be too small for the desired area since some cameras observe outside. I wonder whether it is possible to calculate a bounding box from sparse point clouds (reconstructed by COLMAP), or is it necessary?

[liruilong940607]

If you have access to the point clouds surely it would be good to compute the bbox from that

[AIBluefisher]

I'm implementing this and plan to add it as a feature to nerfacc.