423A35C7/pytorch3d
1
0
Fork 0
mirror of https://github.com/facebookresearch/pytorch3d.git synced 2025-08-02 20:02:49 +08:00
Code Issues Packages Projects Releases Wiki Activity

test_meshes numeric fixes

Browse Source 
Summary:
A couple of tests are failing in open source after my changes yesterday because of numerical issues calculating normals. In particular we have meshes with very few vertices and several faces, where the normals should be zero but end up non-negligible after F.normalize. I have no idea why the different environments produce different results, so that the tests are passing internally.

An example. Consider a mesh with the following faces:
```
tensor([[4, 0, 2],
        [4, 1, 2],
        [3, 1, 0],
        [1, 3, 1],
        [3, 0, 1],
        [4, 0, 0],
        [4, 0, 2]])
```
At vertex 3, there is one zero-area face and there are two other faces, which are back-to-back with each other. This vertex should have zero normal. The open source calculation produces a small but nonzero normal which varies unpredictably with changes in scale/offset, which can cause test failures.

In this diff, the main change is to increase the number of vertices to make this less likely to happen. Also a small change to init_mesh to make it not generate a batch of empty meshes.

Reviewed By: nikhilaravi

Differential Revision: D28220984

fbshipit-source-id: 79fdc62e5f5f8836de5a3a9980cfd6fe44590359
This commit is contained in:
Jeremy Reizenstein 2021-05-07 05:03:10 -07:00 committed by Facebook GitHub Bot
parent 5241b7dd4e
commit e3624b4e9d
1 changed files with 12 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
tests/test_meshes.py
View File

@ -48,7 +48,7 @@ class TestMeshes(TestCaseMixin, unittest.TestCase):
if lists_to_tensors:
# If we define faces/verts with tensors, f/v has to be the
# same for each mesh in the batch.
f = torch.randint(max_f, size=(1,), dtype=torch.int32)
f = torch.randint(1, max_f, size=(1,), dtype=torch.int32)
v = torch.randint(3, high=max_v, size=(1,), dtype=torch.int32)
f = f.repeat(num_meshes)
v = v.repeat(num_meshes)
@ -443,10 +443,10 @@ class TestMeshes(TestCaseMixin, unittest.TestCase):
return Meshes(verts=new_verts_list, faces=new_faces_list)
N = 5
mesh = TestMeshes.init_mesh(N, 10, 100)
mesh = TestMeshes.init_mesh(N, 30, 100, lists_to_tensors=True)
all_v = mesh.verts_packed().size(0)
verts_per_mesh = mesh.num_verts_per_mesh()
for force, deform_shape in itertools.product([0, 1], [(all_v, 3), 3]):
for force, deform_shape in itertools.product([False, True], [(all_v, 3), 3]):
if force:
# force mesh to have computed attributes
mesh._compute_packed(refresh=True)
@ -482,6 +482,7 @@ class TestMeshes(TestCaseMixin, unittest.TestCase):
self.assertClose(
new_mesh.faces_list()[i], new_mesh_naive.faces_list()[i]
)
# check faces and vertex normals
self.assertClose(
new_mesh.verts_normals_list()[i],
@ -491,6 +492,7 @@ class TestMeshes(TestCaseMixin, unittest.TestCase):
self.assertClose(
new_mesh.faces_normals_list()[i],
new_mesh_naive.faces_normals_list()[i],
atol=1e-6,
)
# check padded & packed
@ -544,10 +546,14 @@ class TestMeshes(TestCaseMixin, unittest.TestCase):
atol=1e-6,
)
self.assertClose(
new_mesh.faces_normals_packed(), new_mesh_naive.faces_normals_packed()
new_mesh.faces_normals_packed(),
new_mesh_naive.faces_normals_packed(),
atol=1e-6,
)
self.assertClose(
new_mesh.faces_normals_padded(), new_mesh_naive.faces_normals_padded()
new_mesh.faces_normals_padded(),
new_mesh_naive.faces_normals_padded(),
atol=1e-6,
)
self.assertClose(
new_mesh.faces_areas_packed(), new_mesh_naive.faces_areas_packed()
@ -571,7 +577,7 @@ class TestMeshes(TestCaseMixin, unittest.TestCase):
N = 5
for test in ["tensor", "scalar"]:
for force in (False, True):
mesh = TestMeshes.init_mesh(N, 10, 100)
mesh = TestMeshes.init_mesh(N, 10, 100, lists_to_tensors=True)
if force:
# force mesh to have computed attributes
mesh.verts_packed()