VolumeSampler: expose padding_mode for inside out rendering (#1638)

Summary:
This exposes a setting on VolumeSampler so you can change the padding_mode. This is very useful when using cameras inside a volume that doesn't cover the entire world. By setting the value to `border` you can get much better behavior than `zeros` which causes edge effects for things like the sky. Border emulates infinitely tall buildings instead.

Pull Request resolved: #1638

Test Plan:
Tested with torchdrive

Example before:
![image](https://github.com/facebookresearch/pytorch3d/assets/909104/e99ffb7c-c4ba-40f8-b15c-ad5d1b53f0df)

Example after:
![image](https://github.com/facebookresearch/pytorch3d/assets/909104/f8d9821b-93d5-44b5-b9d4-c1670711ddce)

Reviewed By: MichaelRamamonjisoa

Differential Revision: D49384383

Pulled By: bottler

fbshipit-source-id: 202b526e07320a18944c39a148beec94c0f5d68c

Author: d4l3k

pytorch3d/renderer/implicit/renderer.py

@@ -261,19 +261,28 @@ class VolumeSampler(torch.nn.Module):
     at 3D points sampled along projection rays.
     """
 
-    def __init__(self, volumes: Volumes, sample_mode: str = "bilinear") -> None:
+    def __init__(
+        self,
+        volumes: Volumes,
+        sample_mode: str = "bilinear",
+        padding_mode: str = "zeros",
+    ) -> None:
         """
         Args:
             volumes: An instance of the `Volumes` class representing a
                 batch of volumes that are being rendered.
             sample_mode: Defines the algorithm used to sample the volumetric
                 voxel grid. Can be either "bilinear" or "nearest".
+            padding_mode: How to handle values outside of the volume.
+                One of: zeros, border, reflection
+                See torch.nn.functional.grid_sample for more information.
         """
         super().__init__()
         if not isinstance(volumes, Volumes):
             raise ValueError("'volumes' have to be an instance of the 'Volumes' class.")
         self._volumes = volumes
         self._sample_mode = sample_mode
+        self._padding_mode = padding_mode
 
     def _get_ray_directions_transform(self):
         """
@@ -375,6 +384,7 @@ def forward(
             rays_points_local_flat,
             align_corners=True,
             mode=self._sample_mode,
+            padding_mode=self._padding_mode,
         )
 
         # permute the dimensions & reshape densities after sampling
@@ -392,6 +402,7 @@ def forward(
                 rays_points_local_flat,
                 align_corners=True,
                 mode=self._sample_mode,
+                padding_mode=self._padding_mode,
             )
 
             # permute the dimensions & reshape features after sampling