Fix conv2d with kernel 1x1 stride > 1 (#1868)

Fixes #889

Author: ry

tensorflow/core/kernels/conv_ops.cc

@@ -272,6 +272,7 @@ struct LaunchConvOp<GPUDevice, T> {
     if (use_cudnn) {
       Tensor input = input_param;
       if (filter.dim_size(0) == 1 && filter.dim_size(1) == 1 &&
+          row_stride == 1 && col_stride == 1 &&
           data_format == FORMAT_NHWC) {
         // 1x1 filter, so call cublas directly.
         const uint64 m =

tensorflow/core/kernels/eigen_spatial_convolutions_test.cc

@@ -405,6 +405,55 @@ TEST(EigenSpatialConvolutionsTest, StridedSpatialConvolution) {
   }
 }
 
+
+TEST(EigenSpatialConvolutionsTest, KernelSmallerThanStride) {
+  const int input_depth = 2;
+  const int input_rows = 3;
+  const int input_cols = 3;
+  const int num_batches = 5;
+  const int output_depth = 6;
+  const int patch_rows = 1;
+  const int patch_cols = 1;
+  const int output_rows = 2;
+  const int output_cols = 2;
+
+  Tensor<float, 4> input(input_depth, input_rows, input_cols, num_batches);
+  Tensor<float, 4> kernel(output_depth, input_depth, patch_rows, patch_cols);
+  Tensor<float, 4> result(output_depth, output_rows, output_cols, num_batches);
+  input = input.constant(11.0f) + input.random();
+  kernel = kernel.constant(2.0f) + kernel.random();
+  result.setRandom();
+
+  // Apply a spatial convolution using a 1x1 kernel, valid padding, and a stride
+  // of 2.
+  int stride = 2;
+  result = SpatialConvolution(input, kernel, stride, stride, PADDING_VALID);
+
+  EXPECT_EQ(result.dimension(0), output_depth);
+  EXPECT_EQ(result.dimension(1), output_rows);
+  EXPECT_EQ(result.dimension(2), output_cols);
+  EXPECT_EQ(result.dimension(3), num_batches);
+
+  for (int b = 0; b < num_batches; ++b) {
+    for (int od = 0; od < output_depth; ++od) {
+      for (int i = 0; i < output_rows; ++i) {
+        for (int j = 0; j < output_cols; ++j) {
+          float expected = 0.0f;
+          for (int c = 0; c < patch_cols; ++c) {
+            for (int r = 0; r < patch_rows; ++r) {
+              for (int id = 0; id < input_depth; ++id) {
+                expected += input(id, r + stride * i, c + stride * j, b) *
+                            kernel(od, id, r, c);
+              }
+            }
+          }
+          EigenApprox(result(od, i, j, b), expected);
+        }
+      }
+    }
+  }
+}
+
 TEST(EigenSpatialConvolutionsTest, StridedSpatialConvolutionRowMajor) {
   const int input_depth = 10;
   const int input_rows = 5;

tensorflow/core/kernels/ops_util.cc

@@ -38,11 +38,6 @@ Status Get2dOutputSizeVerbose(const int in_height, const int in_width,
                               int row_stride, int col_stride, Padding padding,
                               int* new_height, int* new_width, int* pad_top,
                               int* pad_bottom, int* pad_left, int* pad_right) {
-  // Cannot have strides larger than the patch size.
-  if (row_stride > filter_height || col_stride > filter_width) {
-    return errors::InvalidArgument(
-        "stride must be less than or equal to kernel size");
-  }
   switch (padding) {
     case Padding::VALID:
       *new_height = ceil((in_height - filter_height + 1.f) /

tensorflow/core/kernels/ops_util_test.cc

@@ -129,12 +129,6 @@ class OpsUtilTest : public ::testing::Test {
   }
 };
 
-// Test stride > ksize fails with INVALID_ARGUMENT.
-TEST_F(OpsUtilTest, Get2dOutputSizeInvalidTest) {
-  padding_struct pad_struct = {{3, 3, 1, 2, 2, 2, SAME}, {3, 3, 1, 1, 1, 1}};
-  VerifyGet2dOutputSizeBoundaries(pad_struct, error::INVALID_ARGUMENT);
-}
-
 TEST_F(OpsUtilTest, Get2dOutputSizeNegativeSizeTest) {
   padding_struct pad_struct = {{1, 1, 3, 3, 1, 1, VALID}, {-1, -1, 0, 0, 0, 0}};
   VerifyGet2dOutputSizeBoundaries(pad_struct, error::INVALID_ARGUMENT);

tensorflow/python/kernel_tests/conv_ops_test.py

@@ -319,6 +319,20 @@ def testConv2D2x2FilterStride1x2(self):
                        strides=[1, 2], padding="VALID",
                        expected=expected_output)
 
+  def testConv2DKernelSmallerThanStrideValid(self):
+    expected_output = [65, 95, 275, 305]
+    self._VerifyValues(tensor_in_sizes=[1, 7, 7, 1],
+                       filter_in_sizes=[2, 2, 1, 1],
+                       strides=[3, 3], padding="VALID",
+                       expected=expected_output)
+
+  def testConv2DKernelSmallerThanStrideSame(self):
+    expected_output = [1, 3, 7, 9]
+    self._VerifyValues(tensor_in_sizes=[1, 3, 3, 1],
+                       filter_in_sizes=[1, 1, 1, 1],
+                       strides=[2, 2], padding="SAME",
+                       expected=expected_output)
+
   # Testing for backprops
   def _RunAndVerifyBackpropInput(self, input_sizes, filter_sizes, output_sizes,
                                  strides, padding, expected, data_format,
@@ -862,21 +876,6 @@ def testShapeFunctionEdgeCases(self):
                                           shape=[21, 20, 3, 2]),
                     strides=[1, 1, 1, 1], padding="SAME")
 
-    # Stride larger than filter.
-    with self.assertRaisesRegexp(ValueError,
-                                 "stride must be less than or equal to filter"):
-      tf.nn.conv2d(tf.placeholder(tf.float32,
-                                          shape=[32, 20, 20, 3]),
-                    tf.placeholder(tf.float32,
-                                          shape=[4, 5, 3, 2]),
-                    strides=[1, 5, 5, 1], padding="SAME")
-    with self.assertRaisesRegexp(ValueError,
-                                 "stride must be less than or equal to filter"):
-      tf.nn.conv2d(tf.placeholder(tf.float32,
-                                          shape=[32, 20, 20, 3]),
-                    tf.placeholder(tf.float32,
-                                          shape=[5, 4, 3, 2]),
-                    strides=[1, 5, 5, 1], padding="SAME")
 
 
 # This is only a very simple test. More comprehensive tests live in

tensorflow/python/kernel_tests/pooling_ops_test.py

@@ -370,6 +370,33 @@ def testDepthwiseMaxPool2x2DepthWindow3(self):
                        expected=[3.0, 6.0, 9.0, 12.0, 15.0, 18.0, 21.0, 24.0],
                        use_gpu=False)
 
+  def testKernelSmallerThanStride(self):
+    for use_gpu in [True, False]:
+        self._VerifyValues(tf.nn.max_pool, input_sizes=[1, 3, 3, 1],
+                           ksize=[1, 1, 1, 1], strides=[1, 2, 2, 1],
+                           padding="SAME",
+                           expected=[1, 3, 7, 9],
+                           use_gpu=use_gpu)
+
+        self._VerifyValues(tf.nn.max_pool, input_sizes=[1, 7, 7, 1],
+                           ksize=[1, 2, 2, 1], strides=[1, 3, 3, 1],
+                           padding="VALID",
+                           expected=[9, 12, 30, 33],
+                           use_gpu=use_gpu)
+
+        self._VerifyValues(tf.nn.avg_pool, input_sizes=[1, 3, 3, 1],
+                           ksize=[1, 1, 1, 1], strides=[1, 2, 2, 1],
+                           padding="SAME",
+                           expected=[1, 3, 7, 9],
+                           use_gpu=use_gpu)
+
+        self._VerifyValues(tf.nn.avg_pool, input_sizes=[1, 7, 7, 1],
+                           ksize=[1, 2, 2, 1], strides=[1, 3, 3, 1],
+                           padding="VALID",
+                           expected=[5, 8, 26, 29],
+                           use_gpu=use_gpu)
+
+
   def _testDepthwiseMaxPoolInvalidConfig(self, in_size, ksize, strides,
                                          error_msg, use_gpu=False):
     t = tf.constant(1.0, shape=in_size)
@@ -885,20 +912,6 @@ def testShapeFunctionEdgeCases(self):
                                         shape=[32, 20, 20, 3]),
                   ksize=[1, 21, 20, 1], strides=[1, 1, 1, 1], padding="SAME")
 
-    # Stride larger than filter.
-    for pool_func in [tf.nn.max_pool, tf.nn.avg_pool,
-                      tf.nn.max_pool_with_argmax]:
-      with self.assertRaisesRegexp(
-          ValueError, "stride must be less than or equal to filter"):
-        pool_func(tf.placeholder(tf.float32,
-                                        shape=[32, 20, 20, 3]),
-                  ksize=[1, 5, 3, 1], strides=[1, 5, 5, 1], padding="SAME")
-      with self.assertRaisesRegexp(
-          ValueError, "stride must be less than or equal to filter"):
-        pool_func(tf.placeholder(tf.float32,
-                                        shape=[32, 20, 20, 3]),
-                  ksize=[1, 3, 5, 1], strides=[1, 5, 5, 1], padding="SAME")
-
 
 def GetMaxPoolFwdTest(input_size, filter_size, strides, padding):
   def Test(self):

tensorflow/python/ops/common_shapes.py

@@ -149,10 +149,6 @@ def get2d_conv_output_size(input_height, input_width, filter_height,
           "filter must not be larger than the input: "
           "Filter: [%sx%s] Input: [%sx%s]"
           % (filter_height, filter_width, input_height, input_width))
-    if row_stride > filter_height or col_stride > filter_width:
-      raise ValueError("stride must be less than or equal to filter size",
-                       "stride: [%sx%s] filter: [%sx%s]"
-                       % (row_stride, col_stride, filter_height, filter_width))
 
     # Compute number of rows in the output, based on the padding.
     if input_height.value is None or filter_height.value is None: