Fix: Pointclouds.inside_box reducing over spatial dimensions.

Summary: As subj. Tests corrected accordingly. Also changed the test to provide a bit better diagnostics.

Reviewed By: bottler

Differential Revision: D32879498

fbshipit-source-id: 0a852e4a13dcb4ca3e54d71c6b263c5d2eeaf4eb

Author: shapovalov

pytorch3d/structures/pointclouds.py

@@ -1176,5 +1176,5 @@ def inside_box(self, box):
             ]
             box = torch.cat(box, 0)
 
-        idx = (points_packed >= box[:, 0]) * (points_packed <= box[:, 1])
-        return idx
+        coord_inside = (points_packed >= box[:, 0]) * (points_packed <= box[:, 1])
+        return coord_inside.all(dim=-1)

tests/common_testing.py

@@ -163,8 +163,17 @@ def assertClose(
         if close:
             return
 
-        diff = backend.abs(input + 0.0 - other)
-        ratio = diff / backend.abs(other)
+        # handle bool case
+        if backend == torch and input.dtype == torch.bool:
+            diff = (input != other).float()
+            ratio = diff
+        if backend == np and input.dtype == bool:
+            diff = (input != other).astype(float)
+            ratio = diff
+        else:
+            diff = backend.abs(input + 0.0 - other)
+            ratio = diff / backend.abs(other)
+
         try_relative = (diff <= atol) | (backend.isfinite(ratio) & (ratio > 0))
         if try_relative.all():
             if backend == np:

tests/test_pointclouds.py

@@ -976,7 +976,9 @@ def test_update_padded(self):
 
     def test_inside_box(self):
         def inside_box_naive(cloud, box_min, box_max):
-            return (cloud >= box_min.view(1, 3)) * (cloud <= box_max.view(1, 3))
+            return ((cloud >= box_min.view(1, 3)) * (cloud <= box_max.view(1, 3))).all(
+                dim=-1
+            )
 
         N, P, C = 5, 100, 4
 
@@ -994,7 +996,7 @@ def inside_box_naive(cloud, box_min, box_max):
         for i, cloud in enumerate(clouds.points_list()):
             within_box_naive.append(inside_box_naive(cloud, box[i, 0], box[i, 1]))
         within_box_naive = torch.cat(within_box_naive, 0)
-        self.assertTrue(within_box.eq(within_box_naive).all())
+        self.assertClose(within_box, within_box_naive)
 
         # box of shape 2x3
         box2 = box[0, :]
@@ -1005,13 +1007,13 @@ def inside_box_naive(cloud, box_min, box_max):
         for cloud in clouds.points_list():
             within_box_naive2.append(inside_box_naive(cloud, box2[0], box2[1]))
         within_box_naive2 = torch.cat(within_box_naive2, 0)
-        self.assertTrue(within_box2.eq(within_box_naive2).all())
+        self.assertClose(within_box2, within_box_naive2)
 
         # box of shape 1x2x3
         box3 = box2.expand(1, 2, 3)
 
         within_box3 = clouds.inside_box(box3)
-        self.assertTrue(within_box2.eq(within_box3).all())
+        self.assertClose(within_box2, within_box3)
 
         # invalid box
         invalid_box = torch.cat(