work with old linalg

Summary: solve and lstsq have moved around in torch. Cope with both.

Reviewed By: patricklabatut

Differential Revision: D29302316

fbshipit-source-id: b34f0b923e90a357f20df359635929241eba6e74

pytorch3d/common/compat.py

@@ -0,0 +1,51 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+from typing import Tuple
+
+import torch
+
+
+"""
+Some functions which depend on PyTorch versions.
+"""
+
+
+def solve(A: torch.Tensor, B: torch.Tensor) -> torch.Tensor:  # pragma: no cover
+    """
+    Like torch.linalg.solve, tries to return X
+    such that AX=B, with A square.
+    """
+    if hasattr(torch.linalg, "solve"):
+        # PyTorch version >= 1.8.0
+        return torch.linalg.solve(A, B)
+
+    return torch.solve(B, A).solution
+
+
+def lstsq(A: torch.Tensor, B: torch.Tensor) -> torch.Tensor:  # pragma: no cover
+    """
+    Like torch.linalg.lstsq, tries to return X
+    such that AX=B.
+    """
+    if hasattr(torch.linalg, "lstsq"):
+        # PyTorch version >= 1.9
+        return torch.linalg.lstsq(A, B).solution
+
+    solution = torch.lstsq(B, A).solution
+    if A.shape[1] < A.shape[0]:
+        return solution[: A.shape[1]]
+    return solution
+
+
+def qr(A: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:  # pragma: no cover
+    """
+    Like torch.linalg.qr.
+    """
+    if hasattr(torch.linalg, "qr"):
+        # PyTorch version >= 1.9
+        return torch.linalg.qr(A)
+    return torch.qr(A)

pytorch3d/transforms/__init__.py

@@ -26,8 +26,8 @@ from .rotation_conversions import (
 )
 from .se3 import se3_exp_map, se3_log_map
 from .so3 import (
-    so3_exponential_map,
     so3_exp_map,
+    so3_exponential_map,
     so3_log_map,
     so3_relative_angle,
     so3_rotation_angle,

pytorch3d/transforms/se3.py

@@ -5,8 +5,9 @@
 # LICENSE file in the root directory of this source tree.
 
 import torch
+from pytorch3d.common.compat import solve
 
-from .so3 import hat, _so3_exp_map, so3_log_map
+from .so3 import _so3_exp_map, hat, so3_log_map
 
 
 def se3_exp_map(log_transform: torch.Tensor, eps: float = 1e-4) -> torch.Tensor:
@@ -173,7 +174,7 @@ def se3_log_map(
     # log_translation is V^-1 @ T
     T = transform[:, 3, :3]
     V = _se3_V_matrix(*_get_se3_V_input(log_rotation), eps=eps)
-    log_translation = torch.linalg.solve(V, T[:, :, None])[:, :, 0]
+    log_translation = solve(V, T[:, :, None])[:, :, 0]
 
     return torch.cat((log_translation, log_rotation), dim=1)
 

pytorch3d/transforms/so3.py

@@ -11,6 +11,7 @@ import torch
 
 from ..transforms import acos_linear_extrapolation
 
+
 HAT_INV_SKEW_SYMMETRIC_TOL = 1e-5
 
 

tests/test_acos_linear_extrapolation.py

@@ -10,6 +10,7 @@ import unittest
 import numpy as np
 import torch
 from common_testing import TestCaseMixin
+from pytorch3d.common.compat import lstsq
 from pytorch3d.transforms import acos_linear_extrapolation
 
 
@@ -64,8 +65,7 @@ class TestAcosLinearExtrapolation(TestCaseMixin, unittest.TestCase):
         bound_t = torch.tensor(bound, device=x.device, dtype=x.dtype)
         # fit a line: slope * x + bias = y
         x_1 = torch.stack([x, torch.ones_like(x)], dim=-1)
-        solution = torch.linalg.lstsq(x_1, y[:, None]).solution
-        slope, bias = solution.view(-1)[:2]
+        slope, bias = lstsq(x_1, y[:, None]).view(-1)[:2]
         desired_slope = (-1.0) / torch.sqrt(1.0 - bound_t ** 2)
         # test that the desired slope is the same as the fitted one
         self.assertClose(desired_slope.view(1), slope.view(1), atol=1e-2)

tests/test_se3.py

@@ -10,13 +10,10 @@ import unittest
 import numpy as np
 import torch
 from common_testing import TestCaseMixin
+from pytorch3d.common.compat import qr
 from pytorch3d.transforms.rotation_conversions import random_rotations
 from pytorch3d.transforms.se3 import se3_exp_map, se3_log_map
-from pytorch3d.transforms.so3 import (
-    so3_exp_map,
-    so3_log_map,
-    so3_rotation_angle,
-)
+from pytorch3d.transforms.so3 import so3_exp_map, so3_log_map, so3_rotation_angle
 
 
 class TestSE3(TestCaseMixin, unittest.TestCase):
@@ -201,7 +198,7 @@ class TestSE3(TestCaseMixin, unittest.TestCase):
         r = [identity, rot180]
         r.extend(
             [
-                torch.qr(identity + torch.randn_like(identity) * 1e-6)[0]
+                qr(identity + torch.randn_like(identity) * 1e-6)[0]
                 + float(i > batch_size // 2) * (0.5 - torch.rand_like(identity)) * 1e-8
                 # this adds random noise to the second half
                 # of the random orthogonal matrices to generate

tests/test_so3.py

@@ -11,6 +11,7 @@ import unittest
 import numpy as np
 import torch
 from common_testing import TestCaseMixin
+from pytorch3d.common.compat import qr
 from pytorch3d.transforms.so3 import (
     hat,
     so3_exp_map,
@@ -46,7 +47,7 @@ class TestSO3(TestCaseMixin, unittest.TestCase):
         # TODO(dnovotny): replace with random_rotation from random_rotation.py
         rot = []
         for _ in range(batch_size):
-            r = torch.qr(torch.randn((3, 3), device=device))[0]
+            r = qr(torch.randn((3, 3), device=device))[0]
             f = torch.randint(2, (3,), device=device, dtype=torch.float32)
             if f.sum() % 2 == 0:
                 f = 1 - f
@@ -142,7 +143,7 @@ class TestSO3(TestCaseMixin, unittest.TestCase):
         # add random rotations and random almost orthonormal matrices
         r.extend(
             [
-                torch.qr(identity + torch.randn_like(identity) * 1e-4)[0]
+                qr(identity + torch.randn_like(identity) * 1e-4)[0]
                 + float(i > batch_size // 2) * (0.5 - torch.rand_like(identity)) * 1e-3
                 # this adds random noise to the second half
                 # of the random orthogonal matrices to generate
@@ -242,7 +243,7 @@ class TestSO3(TestCaseMixin, unittest.TestCase):
         r = [identity, rot180]
         r.extend(
             [
-                torch.qr(identity + torch.randn_like(identity) * 1e-4)[0]
+                qr(identity + torch.randn_like(identity) * 1e-4)[0]
                 for _ in range(batch_size - 2)
             ]
         )