Preserve arbitrary ray batch dimensions in compositing weights

The density and alpha compositing helpers are annotated for arbitrary leading ray batch dimensions, but create identity prefixes from only the first batch axis. The alpha helper also accumulates and slices along dimension 1. Unbatched rays and image-shaped ray batches consequently fail or composite along the wrong axis.

nerfstudio/cameras/rays.py

@@ -141,9 +141,7 @@ class RaySamples(TensorDataclass):
         alphas = 1 - torch.exp(-delta_density)
 
         transmittance = torch.cumsum(delta_density[..., :-1, :], dim=-2)
-        transmittance = torch.cat(
-            [torch.zeros((*transmittance.shape[:1], 1, 1), device=densities.device), transmittance], dim=-2
-        )
+        transmittance = torch.cat([torch.zeros_like(delta_density[..., :1, :]), transmittance], dim=-2)
         transmittance = torch.exp(-transmittance)  # [..., "num_samples"]
 
         weights = alphas * transmittance  # [..., "num_samples"]
@@ -179,10 +177,10 @@ class RaySamples(TensorDataclass):
         """
 
         transmittance = torch.cumprod(
-            torch.cat([torch.ones((*alphas.shape[:1], 1, 1), device=alphas.device), 1.0 - alphas + 1e-7], 1), 1
+            torch.cat([alphas.new_ones((*alphas.shape[:-2], 1, 1)), 1.0 - alphas + 1e-7], dim=-2), dim=-2
         )
 
-        weights = alphas * transmittance[:, :-1, :]
+        weights = alphas * transmittance[..., :-1, :]
         if weights_only:
             return weights
         return weights, transmittance

tests/cameras/test_ray_batch_compositing.py

@@ -0,0 +1,49 @@
+# Copyright 2022 the Regents of the University of California, Nerfstudio Team and contributors. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Volume compositing should preserve all leading ray batch dimensions."""
+
+import pytest
+import torch
+
+from nerfstudio.cameras.rays import Frustums, RaySamples
+
+
+@pytest.mark.parametrize("batch_shape", [(), (2,), (2, 3), (2, 1, 3)])
+@pytest.mark.parametrize("dtype", [torch.float32, torch.float64])
+def test_weight_helpers_preserve_ray_batch_shape(batch_shape, dtype):
+    shape = batch_shape + (4, 1)
+    tau = torch.linspace(0.1, 0.7, 4, dtype=dtype).reshape((1,) * len(batch_shape) + (4, 1)).expand(shape)
+    starts = torch.zeros(shape, dtype=dtype)
+    samples = RaySamples(
+        frustums=Frustums(
+            origins=torch.zeros(batch_shape + (4, 3), dtype=dtype),
+            directions=torch.ones(batch_shape + (4, 3), dtype=dtype),
+            starts=starts,
+            ends=starts + 1,
+            pixel_area=starts + 1,
+        ),
+        deltas=torch.ones_like(tau),
+    )
+    alphas = -torch.expm1(-tau)
+    expected_transmittance = torch.exp(-tau.cumsum(-2) + tau)
+    torch.testing.assert_close(samples.get_weights(tau), alphas * expected_transmittance)
+    weights, transmittance = RaySamples.get_weights_and_transmittance_from_alphas(alphas)
+    reference_transmittance = torch.cat(
+        [alphas.new_ones(batch_shape + (1, 1)), (1 - alphas + 1e-7).cumprod(-2)], dim=-2
+    )
+    assert transmittance.dtype == dtype
+    torch.testing.assert_close(transmittance, reference_transmittance)
+    torch.testing.assert_close(weights, alphas * reference_transmittance[..., :-1, :])
+    torch.testing.assert_close(RaySamples.get_weights_and_transmittance_from_alphas(alphas, weights_only=True), weights)