(breaking) image_size-agnostic GridRaySampler

Summary:
As suggested in #802. By not persisting the _xy_grid buffer, we can allow (in some cases) a model with one image_size to be loaded from a saved model which was trained at a different resolution.

Also avoid persisting _frequencies in HarmonicEmbedding for similar reasons.

BC-break: This will cause load_state_dict, in strict mode, to complain if you try to load an old model with the new code.

Reviewed By: patricklabatut

Differential Revision: D30349234

fbshipit-source-id: d6061d1e51c9f79a78d61a9f732c9a5dfadbbb47

tests/test_raysampling.py

@@ -425,3 +425,23 @@ class TestRaysampling(TestCaseMixin, unittest.TestCase):
             ray_bundle_camera_fix_seed.directions.view(batch_size, -1, 3),
             atol=1e-5,
         )
+
+    def test_load_state(self):
+        # check that we can load the state of one ray sampler into
+        # another with different image size.
+        module1 = NDCGridRaysampler(
+            image_width=20,
+            image_height=30,
+            n_pts_per_ray=40,
+            min_depth=1.2,
+            max_depth=2.3,
+        )
+        module2 = NDCGridRaysampler(
+            image_width=22,
+            image_height=32,
+            n_pts_per_ray=42,
+            min_depth=1.2,
+            max_depth=2.3,
+        )
+        state = module1.state_dict()
+        module2.load_state_dict(state)