fix pts scale, save ply

Summary:
Fix:
* Scaling of point clouds for scalars
* save_ply compatible cat

Reviewed By: nikhilaravi

Differential Revision: D22298609

fbshipit-source-id: abe94a5b64baf325587202d20adfc36912cc1478

pytorch3d/io/ply_io.py

@@ -783,7 +783,11 @@ def save_ply(
         decimal_places: Number of decimal places for saving.
     """
 
-    verts_normals = torch.FloatTensor([]) if verts_normals is None else verts_normals
+    verts_normals = (
+        torch.tensor([], dtype=torch.float32, device=verts.device)
+        if verts_normals is None
+        else verts_normals
+    )
     faces = torch.LongTensor([]) if faces is None else faces
 
     if len(verts) and not (verts.dim() == 2 and verts.size(1) == 3):

pytorch3d/structures/pointclouds.py

@@ -511,7 +511,6 @@ def padded_to_packed_idx(self):
         Returns:
             1D tensor of indices.
         """
-        self._compute_packed()
         if self._padded_to_packed_idx is not None:
             return self._padded_to_packed_idx
         if self._N == 0:
@@ -520,7 +519,7 @@ def padded_to_packed_idx(self):
             self._padded_to_packed_idx = torch.cat(
                 [
                     torch.arange(v, dtype=torch.int64, device=self.device) + i * self._P
-                    for (i, v) in enumerate(self._num_points_per_cloud)
+                    for (i, v) in enumerate(self.num_points_per_cloud())
                 ],
                 dim=0,
             )
@@ -797,7 +796,7 @@ def scale_(self, scale):
             self.
         """
         if not torch.is_tensor(scale):
-            scale = torch.full(len(self), scale)
+            scale = torch.full((len(self),), scale, device=self.device)
         new_points_list = []
         points_list = self.points_list()
         for i, old_points in enumerate(points_list):

tests/test_meshes.py

@@ -510,8 +510,8 @@ def naive_scale_verts(mesh, scale):
 
         N = 5
         for test in ["tensor", "scalar"]:
-            mesh = TestMeshes.init_mesh(N, 10, 100)
-            for force in [0, 1]:
+            for force in (False, True):
+                mesh = TestMeshes.init_mesh(N, 10, 100)
                 if force:
                     # force mesh to have computed attributes
                     mesh.verts_packed()

tests/test_pointclouds.py

@@ -460,7 +460,7 @@ def naive_offset(clouds, offsets_packed):
     def test_scale(self):
         def naive_scale(cloud, scale):
             if not torch.is_tensor(scale):
-                scale = torch.full(len(cloud), scale)
+                scale = torch.full((len(cloud),), scale, device=cloud.device)
             new_points_list = [
                 scale[i] * points.clone()
                 for (i, points) in enumerate(cloud.points_list())
@@ -470,26 +470,37 @@ def naive_scale(cloud, scale):
             )
 
         N = 5
-        clouds = self.init_cloud(N, 100, 10)
-        for force in (False, True):
-            if force:
-                clouds._compute_packed(refresh=True)
-                clouds._compute_padded()
-                clouds.padded_to_packed_idx()
-            scales = torch.rand(N)
-            new_clouds_naive = naive_scale(clouds, scales)
-            new_clouds = clouds.scale(scales)
-            for i in range(N):
-                self.assertClose(
-                    scales[i] * clouds.points_list()[i], new_clouds.points_list()[i]
-                )
-                self.assertClose(
-                    clouds.normals_list()[i], new_clouds_naive.normals_list()[i]
-                )
-                self.assertClose(
-                    clouds.features_list()[i], new_clouds_naive.features_list()[i]
-                )
-            self.assertCloudsEqual(new_clouds, new_clouds_naive)
+        for test in ["tensor", "scalar"]:
+            for force in (False, True):
+                clouds = self.init_cloud(N, 100, 10)
+                if force:
+                    clouds._compute_packed(refresh=True)
+                    clouds._compute_padded()
+                    clouds.padded_to_packed_idx()
+                if test == "tensor":
+                    scales = torch.rand(N)
+                elif test == "scalar":
+                    scales = torch.rand(1)[0].item()
+                new_clouds_naive = naive_scale(clouds, scales)
+                new_clouds = clouds.scale(scales)
+                for i in range(N):
+                    if test == "tensor":
+                        self.assertClose(
+                            scales[i] * clouds.points_list()[i],
+                            new_clouds.points_list()[i],
+                        )
+                    else:
+                        self.assertClose(
+                            scales * clouds.points_list()[i],
+                            new_clouds.points_list()[i],
+                        )
+                    self.assertClose(
+                        clouds.normals_list()[i], new_clouds_naive.normals_list()[i]
+                    )
+                    self.assertClose(
+                        clouds.features_list()[i], new_clouds_naive.features_list()[i]
+                    )
+                self.assertCloudsEqual(new_clouds, new_clouds_naive)
 
     def test_extend_list(self):
         N = 10