Elementwise decoder

Summary: Tensorf does relu or softmax after the density grid. This diff adds the ability to replicate that.

Reviewed By: bottler

Differential Revision: D40023228

fbshipit-source-id: 9f19868cd68460af98ab6e61c7f708158c26dc08

tests/implicitron/test_decoding_functions.py

@@ -1,34 +0,0 @@
-# Copyright (c) Meta Platforms, Inc. and affiliates.
-# All rights reserved.
-#
-# This source code is licensed under the BSD-style license found in the
-# LICENSE file in the root directory of this source tree.
-
-
-import unittest
-
-import torch
-
-from pytorch3d.implicitron.models.implicit_function.decoding_functions import (
-    IdentityDecoder,
-    MLPDecoder,
-)
-from pytorch3d.implicitron.tools.config import expand_args_fields
-
-from tests.common_testing import TestCaseMixin
-
-
-class TestVoxelGrids(TestCaseMixin, unittest.TestCase):
-    def setUp(self):
-        torch.manual_seed(42)
-        expand_args_fields(IdentityDecoder)
-        expand_args_fields(MLPDecoder)
-
-    def test_identity_function(self, in_shape=(33, 4, 1), n_tests=2):
-        """
-        Test that identity function returns its input
-        """
-        func = IdentityDecoder()
-        for _ in range(n_tests):
-            _in = torch.randn(in_shape)
-            assert torch.allclose(func(_in), _in)