matrix_to_quaternion corner case

Summary: Issue #119. The function `sqrt(max(x, 0))` is not convex and has infinite gradient at 0, but 0 is a subgradient at 0. Here we implement it in such a way as to give 0 as the gradient.

Reviewed By: gkioxari

Differential Revision: D24306294

fbshipit-source-id: 48d136faca083babad4d64970be7ea522dbe9e09

Author: bottler

pytorch3d/transforms/rotation_conversions.py

@@ -82,6 +82,17 @@ def _copysign(a, b):
     return torch.where(signs_differ, -a, a)
 
 
+def _sqrt_positive_part(x):
+    """
+    Returns torch.sqrt(torch.max(0, x))
+    but with a zero subgradient where x is 0.
+    """
+    ret = torch.zeros_like(x)
+    positive_mask = x > 0
+    ret[positive_mask] = torch.sqrt(x[positive_mask])
+    return ret
+
+
 def matrix_to_quaternion(matrix):
     """
     Convert rotations given as rotation matrices to quaternions.
@@ -94,14 +105,13 @@ def matrix_to_quaternion(matrix):
     """
     if matrix.size(-1) != 3 or matrix.size(-2) != 3:
         raise ValueError(f"Invalid rotation matrix  shape f{matrix.shape}.")
-    zero = matrix.new_zeros((1,))
     m00 = matrix[..., 0, 0]
     m11 = matrix[..., 1, 1]
     m22 = matrix[..., 2, 2]
-    o0 = 0.5 * torch.sqrt(torch.max(zero, 1 + m00 + m11 + m22))
-    x = 0.5 * torch.sqrt(torch.max(zero, 1 + m00 - m11 - m22))
-    y = 0.5 * torch.sqrt(torch.max(zero, 1 - m00 + m11 - m22))
-    z = 0.5 * torch.sqrt(torch.max(zero, 1 - m00 - m11 + m22))
+    o0 = 0.5 * _sqrt_positive_part(1 + m00 + m11 + m22)
+    x = 0.5 * _sqrt_positive_part(1 + m00 - m11 - m22)
+    y = 0.5 * _sqrt_positive_part(1 - m00 + m11 - m22)
+    z = 0.5 * _sqrt_positive_part(1 - m00 - m11 + m22)
     o1 = _copysign(x, matrix[..., 2, 1] - matrix[..., 1, 2])
     o2 = _copysign(y, matrix[..., 0, 2] - matrix[..., 2, 0])
     o3 = _copysign(z, matrix[..., 1, 0] - matrix[..., 0, 1])

tests/test_rotation_conversions.py

@@ -145,6 +145,20 @@ def test_quaternion_multiplication(self):
         self.assertEqual(ab.shape, ab_from_matrix.shape)
         self.assertTrue(torch.allclose(ab, ab_from_matrix))
 
+    def test_matrix_to_quaternion_corner_case(self):
+        """Check no bad gradients from sqrt(0)."""
+        matrix = torch.eye(3, requires_grad=True)
+        target = torch.Tensor([0.984808, 0, 0.174, 0])
+
+        optimizer = torch.optim.Adam([matrix], lr=0.05)
+        optimizer.zero_grad()
+        q = matrix_to_quaternion(matrix)
+        loss = torch.sum((q - target) ** 2)
+        loss.backward()
+        optimizer.step()
+
+        self.assertClose(matrix, 0.95 * torch.eye(3))
+
     def test_quaternion_application(self):
         """Applying a quaternion is the same as applying the matrix."""
         quaternions = random_quaternions(3, torch.float64, requires_grad=True)