Fix parameters not wrapped with nn.Parameter, antialiasing compatibility

Summary: Some things fail if a parameter is not wraped; in particular, it prevented other tensors moving to GPU.

Reviewed By: bottler

Differential Revision: D40819932

fbshipit-source-id: a23b38ceacd7f0dc131cb0355fef1178e3e2f7fd

Author: shapovalov

pytorch3d/implicitron/models/implicit_function/decoding_functions.py

@@ -75,7 +75,7 @@ class ElementwiseDecoder(DecoderFunctionBase):
         shift: a scalar which is added to the scaled input before performing
             the operation. Defaults to 0.
         operation: which operation to perform on the transformed input. Options are:
-            `relu`, `softplus`, `sigmoid` and `identity`. Defaults to `identity`.
+            `RELU`, `SOFTPLUS`, `SIGMOID` or `IDENTITY`. Defaults to `IDENTITY`.
     """
 
     scale: float = 1
@@ -91,7 +91,7 @@ def __post_init__(self):
             DecoderActivation.IDENTITY,
         ]:
             raise ValueError(
-                "`operation` can only be `relu`, `softplus`, `sigmoid` or identity."
+                "`operation` can only be `RELU`, `SOFTPLUS`, `SIGMOID` or `IDENTITY`."
             )
 
     def forward(
@@ -165,22 +165,18 @@ class MLPWithInputSkips(Configurable, torch.nn.Module):
     def __post_init__(self):
         super().__init__()
 
-        if self.last_activation not in [
-            DecoderActivation.RELU,
-            DecoderActivation.SOFTPLUS,
-            DecoderActivation.SIGMOID,
-            DecoderActivation.IDENTITY,
-        ]:
+        try:
+            last_activation = {
+                DecoderActivation.RELU: torch.nn.ReLU(True),
+                DecoderActivation.SOFTPLUS: torch.nn.Softplus(),
+                DecoderActivation.SIGMOID: torch.nn.Sigmoid(),
+                DecoderActivation.IDENTITY: torch.nn.Identity(),
+            }[self.last_activation]
+        except KeyError as e:
             raise ValueError(
-                "`last_activation` can only be `relu`,"
-                " `softplus`, `sigmoid` or identity."
-            )
-        last_activation = {
-            DecoderActivation.RELU: torch.nn.ReLU(True),
-            DecoderActivation.SOFTPLUS: torch.nn.Softplus(),
-            DecoderActivation.SIGMOID: torch.nn.Sigmoid(),
-            DecoderActivation.IDENTITY: torch.nn.Identity(),
-        }[self.last_activation]
+                "`last_activation` can only be `RELU`,"
+                " `SOFTPLUS`, `SIGMOID` or `IDENTITY`."
+            ) from e
 
         layers = []
         skip_affine_layers = []

pytorch3d/implicitron/models/implicit_function/voxel_grid.py

@@ -15,9 +15,12 @@
 
 """
 
+import warnings
 from collections.abc import Mapping
 from dataclasses import dataclass, field
-from typing import Callable, ClassVar, Dict, Iterator, List, Optional, Tuple, Type
+
+from distutils.version import LooseVersion
+from typing import Any, Callable, ClassVar, Dict, Iterator, List, Optional, Tuple, Type
 
 import torch
 from omegaconf import DictConfig
@@ -67,7 +70,9 @@ class VoxelGridBase(ReplaceableBase, torch.nn.Module):
     padding: str = "zeros"
     mode: str = "bilinear"
     n_features: int = 1
-    resolution_changes: Dict[int, List[int]] = field(
+    # return the line below once we drop OmegaConf 2.1 support
+    # resolution_changes: Dict[int, List[int]] = field(
+    resolution_changes: Dict[int, Any] = field(
         default_factory=lambda: {0: [128, 128, 128]}
     )
 
@@ -212,6 +217,13 @@ def change_resolution(
                 + "| 'bicubic' | 'linear' | 'area' | 'nearest-exact'"
             )
 
+        interpolate_has_antialias = LooseVersion(torch.__version__) >= "1.11"
+
+        if antialias and not interpolate_has_antialias:
+            warnings.warn("Antialiased interpolation requires PyTorch 1.11+; ignoring")
+
+        interp_kwargs = {"antialias": antialias} if interpolate_has_antialias else {}
+
         def change_individual_resolution(tensor, wanted_resolution):
             if mode == "linear":
                 n_dim = len(wanted_resolution)
@@ -223,8 +235,8 @@ def change_individual_resolution(tensor, wanted_resolution):
                 size=wanted_resolution,
                 mode=new_mode,
                 align_corners=align_corners,
-                antialias=antialias,
                 recompute_scale_factor=False,
+                **interp_kwargs,
             )
 
         if epoch is not None:
@@ -880,7 +892,14 @@ def set_voxel_grid_parameters(self, params: VoxelGridValuesBase) -> None:
         """
         if self.hold_voxel_grid_as_parameters:
             # pyre-ignore [16]
-            self.params = torch.nn.ParameterDict(vars(params))
+            # Nones are converted to empty tensors by Parameter()
+            self.params = torch.nn.ParameterDict(
+                {
+                    k: torch.nn.Parameter(val)
+                    for k, val in vars(params).items()
+                    if val is not None
+                }
+            )
         else:
             # Torch Module to hold parameters since they can only be registered
             # at object level.
@@ -1011,7 +1030,11 @@ def crop_self(self, min_point: torch.Tensor, max_point: torch.Tensor) -> None:
         )
         # pyre-ignore [16]
         self.params = torch.nn.ParameterDict(
-            {k: v for k, v in vars(grid_values).items()}
+            {
+                k: torch.nn.Parameter(val)
+                for k, val in vars(grid_values).items()
+                if val is not None
+            }
         )
         # New center of voxel grid is the middle point between max and min points.
         self.translation = tuple((max_point + min_point) / 2)

pytorch3d/implicitron/models/implicit_function/voxel_grid_implicit_function.py

@@ -527,10 +527,10 @@ def _get_scaffold(self, epoch: int) -> bool:
         return False
 
     @classmethod
-    def decoder_density_tweak_args(cls, type, args: DictConfig) -> None:
+    def decoder_density_tweak_args(cls, type_, args: DictConfig) -> None:
         args.pop("input_dim", None)
 
-    def create_decoder_density_impl(self, type, args: DictConfig) -> None:
+    def create_decoder_density_impl(self, type_, args: DictConfig) -> None:
         """
         Decoding functions come after harmonic embedding and voxel grid. In order to not
         calculate the input dimension of the decoder in the config file this function
@@ -548,18 +548,18 @@ def create_decoder_density_impl(self, type, args: DictConfig) -> None:
             embedder_args["append_input"],
         )
 
-        cls = registry.get(DecoderFunctionBase, type)
+        cls = registry.get(DecoderFunctionBase, type_)
         need_input_dim = any(field.name == "input_dim" for field in fields(cls))
         if need_input_dim:
             self.decoder_density = cls(input_dim=input_dim, **args)
         else:
             self.decoder_density = cls(**args)
 
     @classmethod
-    def decoder_color_tweak_args(cls, type, args: DictConfig) -> None:
+    def decoder_color_tweak_args(cls, type_, args: DictConfig) -> None:
         args.pop("input_dim", None)
 
-    def create_decoder_color_impl(self, type, args: DictConfig) -> None:
+    def create_decoder_color_impl(self, type_, args: DictConfig) -> None:
         """
         Decoding functions come after harmonic embedding and voxel grid. In order to not
         calculate the input dimension of the decoder in the config file this function
@@ -587,7 +587,7 @@ def create_decoder_color_impl(self, type, args: DictConfig) -> None:
 
         input_dim = input_dim0 + input_dim1
 
-        cls = registry.get(DecoderFunctionBase, type)
+        cls = registry.get(DecoderFunctionBase, type_)
         need_input_dim = any(field.name == "input_dim" for field in fields(cls))
         if need_input_dim:
             self.decoder_color = cls(input_dim=input_dim, **args)