C++ API: torch::nn::BatchNorm{2,3}d (#28936)

Summary:
Add torch::nn::BatchNorm{2,3}d module and functional support for the C++ API.

Related Issue: https://github.com/pytorch/pytorch/issues/25883 #28176

Reviewer: yf225
Pull Request resolved: https://github.com/pytorch/pytorch/pull/28936

Differential Revision: D18274584

Pulled By: yf225

fbshipit-source-id: 3784eee9f8947f6c7c9f1699544a3d36a1a019b7
This commit is contained in:
nuka137
2019-11-01 17:48:47 -07:00
committed by Facebook Github Bot
parent 23193c155f
commit a68c1e109e
6 changed files with 220 additions and 2 deletions
+60
View File
@@ -1362,6 +1362,66 @@ TEST_F(FunctionalTest, BatchNorm1dDefaultOptions) {
ASSERT_TRUE(output.allclose(expected));
}
TEST_F(FunctionalTest, BatchNorm2d) {
int num_features = 5;
double eps = 1e-05;
double momentum = 0.1;
auto input = torch::randn({2, num_features, 4, 4});
auto mean = torch::randn(num_features);
auto variance = torch::rand(num_features);
auto weight = torch::ones({num_features});
auto bias = torch::zeros({num_features});
auto output = F::batch_norm(
input, mean, variance,
BatchNormOptions().weight(weight).bias(bias).momentum(momentum).eps(eps),
/*training=*/false);
auto expected = torch::transpose((torch::transpose(input, 1, 3) - mean) / torch::sqrt(variance + eps), 1, 3);
ASSERT_TRUE(output.allclose(expected));
}
TEST_F(FunctionalTest, BatchNorm2dDefaultOptions) {
int num_features = 5;
double eps = 1e-05;
auto input = torch::randn({2, num_features, 4, 4});
auto mean = torch::randn(num_features);
auto variance = torch::rand(num_features);
auto output = F::batch_norm(input, mean, variance);
auto expected = torch::transpose((torch::transpose(input, 1, 3) - mean) / torch::sqrt(variance + eps), 1, 3);
ASSERT_TRUE(output.allclose(expected));
}
TEST_F(FunctionalTest, BatchNorm3d) {
int num_features = 5;
double eps = 1e-05;
double momentum = 0.1;
auto input = torch::randn({2, num_features, 2, 2, 2});
auto mean = torch::randn(num_features);
auto variance = torch::rand(num_features);
auto weight = torch::ones({num_features});
auto bias = torch::zeros({num_features});
auto output = F::batch_norm(
input, mean, variance,
BatchNormOptions().weight(weight).bias(bias).momentum(momentum).eps(eps),
/*training=*/false);
auto expected = torch::transpose((torch::transpose(input, 1, 4) - mean) / torch::sqrt(variance + eps), 1, 4);
ASSERT_TRUE(output.allclose(expected));
}
TEST_F(FunctionalTest, BatchNorm3dDefaultOptions) {
int num_features = 5;
double eps = 1e-05;
auto input = torch::randn({2, num_features, 2, 2, 2});
auto mean = torch::randn(num_features);
auto variance = torch::rand(num_features);
auto output = F::batch_norm(input, mean, variance);
auto expected = torch::transpose((torch::transpose(input, 1, 4) - mean) / torch::sqrt(variance + eps), 1, 4);
ASSERT_TRUE(output.allclose(expected));
}
TEST_F(FunctionalTest, Interpolate) {
{
// 1D interpolation