Enable __getitem__ for Cameras to return an instance of Cameras

Summary:
Added a custom `__getitem__` method to `CamerasBase` which returns an instance of the appropriate camera instead of the `TensorAccessor` class.

Long term we should deprecate the `TensorAccessor` and the `__getitem__` method on `TensorProperties`

FB: In the next diff I will update the uses of `select_cameras` in implicitron.

Reviewed By: bottler

Differential Revision: D33185885

fbshipit-source-id: c31995d0eb126981e91ba61a6151d5404b263f67

pytorch3d/renderer/cameras.py

@@ -75,6 +75,10 @@ class CamerasBase(TensorProperties):
     boolean argument of the function.
     """
 
+    # Used in __getitem__ to index the relevant fields
+    # When creating a new camera, this should be set in the __init__
+    _FIELDS: Tuple = ()
+
     def get_projection_transform(self):
         """
         Calculate the projective transformation matrix.
@@ -362,6 +366,55 @@ def get_image_size(self):
         """
         return self.image_size if hasattr(self, "image_size") else None
 
+    def __getitem__(
+        self, index: Union[int, List[int], torch.LongTensor]
+    ) -> "CamerasBase":
+        """
+        Override for the __getitem__ method in TensorProperties which needs to be
+        refactored.
+
+        Args:
+            index: an int/list/long tensor used to index all the fields in the cameras given by
+                self._FIELDS.
+        Returns:
+            if `index` is an index int/list/long tensor return an instance of the current
+            cameras class with only the values at the selected index.
+        """
+
+        kwargs = {}
+
+        if not isinstance(index, (int, list, torch.LongTensor)):
+            msg = "Invalid index type, expected int, List[int] or torch.LongTensor; got %r"
+            raise ValueError(msg % type(index))
+
+        if isinstance(index, int):
+            index = [index]
+
+        if max(index) >= len(self):
+            raise ValueError(f"Index {max(index)} is out of bounds for select cameras")
+
+        for field in self._FIELDS:
+            val = getattr(self, field, None)
+            if val is None:
+                continue
+
+            # e.g. "in_ndc" is set as attribute "_in_ndc" on the class
+            # but provided as "in_ndc" on initialization
+            if field.startswith("_"):
+                field = field[1:]
+
+            if isinstance(val, (str, bool)):
+                kwargs[field] = val
+            elif isinstance(val, torch.Tensor):
+                # In the init, all inputs will be converted to
+                # tensors before setting as attributes
+                kwargs[field] = val[index]
+            else:
+                raise ValueError(f"Field {field} type is not supported for indexing")
+
+        kwargs["device"] = self.device
+        return self.__class__(**kwargs)
+
 
 ############################################################
 #             Field of View Camera Classes                 #
@@ -434,6 +487,18 @@ class FoVPerspectiveCameras(CamerasBase):
     for rasterization.
     """
 
+    # For __getitem__
+    _FIELDS = (
+        "K",
+        "znear",
+        "zfar",
+        "aspect_ratio",
+        "fov",
+        "R",
+        "T",
+        "degrees",
+    )
+
     def __init__(
         self,
         znear=1.0,
@@ -590,7 +655,7 @@ def unproject_points(
         xy_depth: torch.Tensor,
         world_coordinates: bool = True,
         scaled_depth_input: bool = False,
-        **kwargs
+        **kwargs,
     ) -> torch.Tensor:
         """>!
         FoV cameras further allow for passing depth in world units
@@ -681,6 +746,20 @@ class FoVOrthographicCameras(CamerasBase):
     The definition of the parameters follow the OpenGL orthographic camera.
     """
 
+    # For __getitem__
+    _FIELDS = (
+        "K",
+        "znear",
+        "zfar",
+        "R",
+        "T",
+        "max_y",
+        "min_y",
+        "max_x",
+        "min_x",
+        "scale_xyz",
+    )
+
     def __init__(
         self,
         znear=1.0,
@@ -819,7 +898,7 @@ def unproject_points(
         xy_depth: torch.Tensor,
         world_coordinates: bool = True,
         scaled_depth_input: bool = False,
-        **kwargs
+        **kwargs,
     ) -> torch.Tensor:
         """>!
         FoV cameras further allow for passing depth in world units
@@ -907,6 +986,17 @@ class PerspectiveCameras(CamerasBase):
     If parameters are specified in screen space, `in_ndc` must be set to False.
     """
 
+    # For __getitem__
+    _FIELDS = (
+        "K",
+        "R",
+        "T",
+        "focal_length",
+        "principal_point",
+        "_in_ndc",  # arg is in_ndc but attribute set as _in_ndc
+        "image_size",
+    )
+
     def __init__(
         self,
         focal_length=1.0,
@@ -1007,7 +1097,7 @@ def unproject_points(
         xy_depth: torch.Tensor,
         world_coordinates: bool = True,
         from_ndc: bool = False,
-        **kwargs
+        **kwargs,
     ) -> torch.Tensor:
         """
         Args:
@@ -1126,6 +1216,17 @@ class OrthographicCameras(CamerasBase):
     If parameters are specified in screen space, `in_ndc` must be set to False.
     """
 
+    # For __getitem__
+    _FIELDS = (
+        "K",
+        "R",
+        "T",
+        "focal_length",
+        "principal_point",
+        "_in_ndc",
+        "image_size",
+    )
+
     def __init__(
         self,
         focal_length=1.0,
@@ -1225,7 +1326,7 @@ def unproject_points(
         xy_depth: torch.Tensor,
         world_coordinates: bool = True,
         from_ndc: bool = False,
-        **kwargs
+        **kwargs,
     ) -> torch.Tensor:
         """
         Args:

pytorch3d/renderer/utils.py

@@ -155,7 +155,7 @@ def __getitem__(self, index: Union[int, slice]) -> TensorAccessor:
         Returns:
             if `index` is an index int/slice return a TensorAccessor class
             with getattribute/setattribute methods which return/update the value
-            at the index in the original camera.
+            at the index in the original class.
         """
         if isinstance(index, (int, slice)):
             return TensorAccessor(class_object=self, index=index)

tests/test_cameras.py

@@ -783,18 +783,53 @@ def test_camera_class_init(self):
         self.assertTrue(cam.znear.shape == (2,))
         self.assertTrue(cam.zfar.shape == (2,))
 
-        # update znear element 1
-        cam[1].znear = 20.0
-        self.assertTrue(cam.znear[1] == 20.0)
-
-        # Get item and get value
-        c0 = cam[0]
-        self.assertTrue(c0.zfar == 100.0)
-
         # Test to
         new_cam = cam.to(device=device)
         self.assertTrue(new_cam.device == device)
 
+    def test_getitem(self):
+        R_matrix = torch.randn((6, 3, 3))
+        cam = FoVPerspectiveCameras(znear=10.0, zfar=100.0, R=R_matrix)
+
+        # Check get item returns an instance of the same class
+        # with all the same keys
+        c0 = cam[0]
+        self.assertTrue(isinstance(c0, FoVPerspectiveCameras))
+        self.assertEqual(cam.__dict__.keys(), c0.__dict__.keys())
+
+        # Check all fields correct in get item with int index
+        self.assertEqual(len(c0), 1)
+        self.assertClose(c0.zfar, torch.tensor([100.0]))
+        self.assertClose(c0.znear, torch.tensor([10.0]))
+        self.assertClose(c0.R, R_matrix[0:1, ...])
+        self.assertEqual(c0.device, torch.device("cpu"))
+
+        # Check list(int) index
+        c012 = cam[[0, 1, 2]]
+        self.assertEqual(len(c012), 3)
+        self.assertClose(c012.zfar, torch.tensor([100.0] * 3))
+        self.assertClose(c012.znear, torch.tensor([10.0] * 3))
+        self.assertClose(c012.R, R_matrix[0:3, ...])
+
+        # Check torch.LongTensor index
+        index = torch.tensor([1, 3, 5], dtype=torch.int64)
+        c135 = cam[index]
+        self.assertEqual(len(c135), 3)
+        self.assertClose(c135.zfar, torch.tensor([100.0] * 3))
+        self.assertClose(c135.znear, torch.tensor([10.0] * 3))
+        self.assertClose(c135.R, R_matrix[[1, 3, 5], ...])
+
+        # Check errors with get item
+        with self.assertRaisesRegex(ValueError, "out of bounds"):
+            cam[6]
+
+        with self.assertRaisesRegex(ValueError, "Invalid index type"):
+            cam[slice(0, 1)]
+
+        with self.assertRaisesRegex(ValueError, "Invalid index type"):
+            index = torch.tensor([1, 3, 5], dtype=torch.float32)
+            cam[index]
+
     def test_get_full_transform(self):
         cam = FoVPerspectiveCameras()
         T = torch.tensor([0.0, 0.0, 1.0]).view(1, -1)
@@ -919,6 +954,30 @@ def test_perspective_type(self):
         self.assertFalse(cam.is_perspective())
         self.assertEqual(cam.get_znear(), 1.0)
 
+    def test_getitem(self):
+        R_matrix = torch.randn((6, 3, 3))
+        scale = torch.tensor([[1.0, 1.0, 1.0]], requires_grad=True)
+        cam = FoVOrthographicCameras(
+            znear=10.0, zfar=100.0, R=R_matrix, scale_xyz=scale
+        )
+
+        # Check get item returns an instance of the same class
+        # with all the same keys
+        c0 = cam[0]
+        self.assertTrue(isinstance(c0, FoVOrthographicCameras))
+        self.assertEqual(cam.__dict__.keys(), c0.__dict__.keys())
+
+        # Check torch.LongTensor index
+        index = torch.tensor([1, 3, 5], dtype=torch.int64)
+        c135 = cam[index]
+        self.assertEqual(len(c135), 3)
+        self.assertClose(c135.zfar, torch.tensor([100.0] * 3))
+        self.assertClose(c135.znear, torch.tensor([10.0] * 3))
+        self.assertClose(c135.min_x, torch.tensor([-1.0] * 3))
+        self.assertClose(c135.max_x, torch.tensor([1.0] * 3))
+        self.assertClose(c135.R, R_matrix[[1, 3, 5], ...])
+        self.assertClose(c135.scale_xyz, scale.expand(3, -1))
+
 
 ############################################################
 #                Orthographic Camera                       #
@@ -976,6 +1035,30 @@ def test_perspective_type(self):
         self.assertFalse(cam.is_perspective())
         self.assertIsNone(cam.get_znear())
 
+    def test_getitem(self):
+        R_matrix = torch.randn((6, 3, 3))
+        principal_point = torch.randn((6, 2, 1))
+        focal_length = 5.0
+        cam = OrthographicCameras(
+            R=R_matrix,
+            focal_length=focal_length,
+            principal_point=principal_point,
+        )
+
+        # Check get item returns an instance of the same class
+        # with all the same keys
+        c0 = cam[0]
+        self.assertTrue(isinstance(c0, OrthographicCameras))
+        self.assertEqual(cam.__dict__.keys(), c0.__dict__.keys())
+
+        # Check torch.LongTensor index
+        index = torch.tensor([1, 3, 5], dtype=torch.int64)
+        c135 = cam[index]
+        self.assertEqual(len(c135), 3)
+        self.assertClose(c135.focal_length, torch.tensor([5.0] * 3))
+        self.assertClose(c135.R, R_matrix[[1, 3, 5], ...])
+        self.assertClose(c135.principal_point, principal_point[[1, 3, 5], ...])
+
 
 ############################################################
 #                Perspective Camera                        #
@@ -1027,3 +1110,30 @@ def test_perspective_type(self):
         cam = PerspectiveCameras(focal_length=5.0, principal_point=((2.5, 2.5),))
         self.assertTrue(cam.is_perspective())
         self.assertIsNone(cam.get_znear())
+
+    def test_getitem(self):
+        R_matrix = torch.randn((6, 3, 3))
+        principal_point = torch.randn((6, 2, 1))
+        focal_length = 5.0
+        cam = PerspectiveCameras(
+            R=R_matrix,
+            focal_length=focal_length,
+            principal_point=principal_point,
+        )
+
+        # Check get item returns an instance of the same class
+        # with all the same keys
+        c0 = cam[0]
+        self.assertTrue(isinstance(c0, PerspectiveCameras))
+        self.assertEqual(cam.__dict__.keys(), c0.__dict__.keys())
+
+        # Check torch.LongTensor index
+        index = torch.tensor([1, 3, 5], dtype=torch.int64)
+        c135 = cam[index]
+        self.assertEqual(len(c135), 3)
+        self.assertClose(c135.focal_length, torch.tensor([5.0] * 3))
+        self.assertClose(c135.R, R_matrix[[1, 3, 5], ...])
+        self.assertClose(c135.principal_point, principal_point[[1, 3, 5], ...])
+
+        # Check in_ndc is handled correctly
+        self.assertEqual(cam._in_ndc, c0._in_ndc)