Match gridbatch and grid API (#282)

Changed some things in grid, now changing them in gridbatch too.

1. Rigorous naming of coordinates spaces (world means world, voxel means
ijk, index means linear offset). Updated methods to reflect this.
2. Moved load/save functions to from_nanovdb, save_nanovdb which
conforms with `Grid` and `GaussianSplat3d` (`from_ply`/`save_ply`).
3. Renamed `read/write_to/from_dense_cmajor/minor` to
`inject_to/from_dense_cmajor/minor` because these are just injection
functions over dense grids and it conforms with `inject_from_ijk`.

Signed-off-by: Francis Williams <francis@fwilliams.info>

Author: fwilliams

examples/wip/structure_prediction_net.py

@@ -4,18 +4,16 @@
 
 import os
 
+import fvdb.nn as fvnn
 import numpy as np
+import point_cloud_utils as pcu
+import polyscope as ps
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
-
-import point_cloud_utils as pcu
-import polyscope as ps
-
 import torch_cudamanaged
 
 import fvdb
-import fvdb.nn as fvnn
 
 torch.backends.cudnn.deterministic = True
 np.random.seed(42)
@@ -125,7 +123,7 @@ def normalize_pts(xyz: np.ndarray):
 
 def visualize_grid(grid: fvdb.GridBatch):
     ps.init()
-    xyz = grid.grid_to_world(grid.ijk.float())
+    xyz = grid.voxel_to_world(grid.ijk.float())
     for batch_idx in range(grid.grid_count):
         pts = xyz[batch_idx].jdata.cpu().numpy()
         ps.register_point_cloud(f"grid_{batch_idx}", pts, radius=0.0025)

fvdb/__init__.py

@@ -75,14 +75,8 @@ def _parse_device_string(device_or_device_string: str | torch.device) -> torch.d
     jempty,
 )
 
-# Import GridBatch and gridbatch_from_* functions from grid_batch.py
-from .grid_batch import (
-    GridBatch,
-    load_gridbatch,
-    save_gridbatch,
-)
-
-
+# Import GridBatch and Grid
+from .grid_batch import GridBatch
 from .grid import Grid