-
Notifications
You must be signed in to change notification settings - Fork 1.3k
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Replace
torch.det()
with manual implementation for 3x3 matrix
Summary: # Background There is an unstable error during training (it can happen after several minutes or after several hours). The error is connected to `torch.det()` function in `_check_valid_rotation_matrix()`. if I remove the function `torch.det()` in `_check_valid_rotation_matrix()` or remove the whole functions `_check_valid_rotation_matrix()` the error is disappeared (D29555876). # Solution Replace `torch.det()` with manual implementation for 3x3 matrix. Reviewed By: patricklabatut Differential Revision: D29655924 fbshipit-source-id: 41bde1119274a705ab849751ece28873d2c45155
- Loading branch information
1 parent
2f668ec
commit bcee361
Showing
3 changed files
with
89 additions
and
1 deletion.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,31 @@ | ||
# 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. | ||
|
||
|
||
import torch | ||
|
||
|
||
def _safe_det_3x3(t: torch.Tensor): | ||
""" | ||
Fast determinant calculation for a batch of 3x3 matrices. | ||
Note, result of this function might not be the same as `torch.det()`. | ||
The differences might be in the last significant digit. | ||
Args: | ||
t: Tensor of shape (N, 3, 3). | ||
Returns: | ||
Tensor of shape (N) with determinants. | ||
""" | ||
|
||
det = ( | ||
t[..., 0, 0] * (t[..., 1, 1] * t[..., 2, 2] - t[..., 1, 2] * t[..., 2, 1]) | ||
- t[..., 0, 1] * (t[..., 1, 0] * t[..., 2, 2] - t[..., 2, 0] * t[..., 1, 2]) | ||
+ t[..., 0, 2] * (t[..., 1, 0] * t[..., 2, 1] - t[..., 2, 0] * t[..., 1, 1]) | ||
) | ||
|
||
return det |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,56 @@ | ||
# 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. | ||
|
||
|
||
import unittest | ||
|
||
import numpy as np | ||
import torch | ||
from common_testing import TestCaseMixin | ||
from pytorch3d.common.workaround import _safe_det_3x3 | ||
|
||
|
||
class TestSafeDet3x3(TestCaseMixin, unittest.TestCase): | ||
def setUp(self) -> None: | ||
super().setUp() | ||
torch.manual_seed(42) | ||
np.random.seed(42) | ||
|
||
def _test_det_3x3(self, batch_size, device): | ||
t = torch.rand((batch_size, 3, 3), dtype=torch.float32, device=device) | ||
actual_det = _safe_det_3x3(t) | ||
expected_det = t.det() | ||
self.assertClose(actual_det, expected_det, atol=1e-7) | ||
|
||
def test_empty_batch(self): | ||
self._test_det_3x3(0, torch.device("cpu")) | ||
self._test_det_3x3(0, torch.device("cuda:0")) | ||
|
||
def test_manual(self): | ||
t = torch.Tensor( | ||
[ | ||
[[1, 0, 0], [0, 1, 0], [0, 0, 1]], | ||
[[2, -5, 3], [0, 7, -2], [-1, 4, 1]], | ||
[[6, 1, 1], [4, -2, 5], [2, 8, 7]], | ||
] | ||
).to(dtype=torch.float32) | ||
expected_det = torch.Tensor([1, 41, -306]).to(dtype=torch.float32) | ||
self.assertClose(_safe_det_3x3(t), expected_det) | ||
|
||
device_cuda = torch.device("cuda:0") | ||
self.assertClose( | ||
_safe_det_3x3(t.to(device=device_cuda)), expected_det.to(device=device_cuda) | ||
) | ||
|
||
def test_regression(self): | ||
tries = 32 | ||
device_cpu = torch.device("cpu") | ||
device_cuda = torch.device("cuda:0") | ||
batch_sizes = np.random.randint(low=1, high=128, size=tries) | ||
|
||
for batch_size in batch_sizes: | ||
self._test_det_3x3(batch_size, device_cpu) | ||
self._test_det_3x3(batch_size, device_cuda) |