commit 1605fc9bda5e77a314c13d101f924971d3f52717
author Aron Virginas-Tar <Aron.Virginas-Tar@arm.com> Tue Sep 17 12:58:22 2019 +0100
committer David Monahan <david.monahan@arm.com> Wed Jan 15 15:56:58 2020 +0000
tree dffc56aa64ecff3ea35b76d8b9d45c761e3e06f4
parent 4b7adae342658d81702301d8094b9212d73338ff

IVGCVSW-3879 Fix output shape inference formula for TransposeConvolution2d

Signed-off-by: Aron Virginas-Tar <Aron.Virginas-Tar@arm.com>
Change-Id: I766f4297b9daa26edacc2079fe62a083ba2fa68f

src/armnn/layers/TransposeConvolution2dLayer.cpp

diff --git a/src/armnn/layers/TransposeConvolution2dLayer.cpp b/src/armnn/layers/TransposeConvolution2dLayer.cpp
index 77a333d..7bd2f3b 100644
--- a/src/armnn/layers/TransposeConvolution2dLayer.cpp
+++ b/src/armnn/layers/TransposeConvolution2dLayer.cpp
@@ -66,7 +66,7 @@
 
     DataLayoutIndexed dataLayoutIndex(m_Param.m_DataLayout);
 
-    const unsigned int batches  = inputShape[0];
+    const unsigned int batches = inputShape[0];
 
     const unsigned int wInput = inputShape[dataLayoutIndex.GetWidthIndex()];
     const unsigned int hInput = inputShape[dataLayoutIndex.GetHeightIndex()];
@@ -74,20 +74,18 @@
     const unsigned int wKernel = kernelShape[dataLayoutIndex.GetWidthIndex()];
     const unsigned int hKernel = kernelShape[dataLayoutIndex.GetHeightIndex()];
 
-    const unsigned int wStridedInput = 1u + m_Param.m_StrideX * (wInput - 1);
-    const unsigned int hStridedInput = 1u + m_Param.m_StrideY * (hInput - 1);
+    unsigned int wPadding = m_Param.m_PadLeft + m_Param.m_PadRight;
+    unsigned int hPadding = m_Param.m_PadTop + m_Param.m_PadBottom;
 
-    const unsigned int wPaddedOutput = wStridedInput + wKernel - (wKernel % 2);
-    const unsigned int hPaddedOutput = hStridedInput + hKernel - (hKernel % 2);
-
-    unsigned int wOutput = wPaddedOutput - (m_Param.m_PadLeft + m_Param.m_PadRight);
-    unsigned int hOutput = hPaddedOutput - (m_Param.m_PadTop + m_Param.m_PadBottom);
+    unsigned int wOutput = (wInput - 1) * m_Param.m_StrideX + wKernel - wPadding;
+    unsigned int hOutput = (hInput - 1) * m_Param.m_StrideY + hKernel - hPadding;
 
     unsigned int kernelElements = kernelShape[0] * kernelShape[dataLayoutIndex.GetChannelsIndex()];
-    unsigned int inputElements = batches * inputShape[dataLayoutIndex.GetChannelsIndex()];
+    unsigned int inputElements  = batches * inputShape[dataLayoutIndex.GetChannelsIndex()];
 
     BOOST_ASSERT_MSG(inputElements != 0, "Invalid number of input elements");
     BOOST_ASSERT_MSG(kernelElements % inputElements == 0, "Invalid number of elements");
+
     unsigned int channels =  kernelElements / inputElements;
 
     TensorShape tensorShape = m_Param.m_DataLayout == armnn::DataLayout::NHWC ?

src/armnnTfLiteParser/test/TransposeConv.cpp

diff --git a/src/armnnTfLiteParser/test/TransposeConv.cpp b/src/armnnTfLiteParser/test/TransposeConv.cpp
index 53212b5..46b02ac 100644
--- a/src/armnnTfLiteParser/test/TransposeConv.cpp
+++ b/src/armnnTfLiteParser/test/TransposeConv.cpp
@@ -83,7 +83,7 @@
                             "outputs": [ 3 ],
                             "builtin_options_type": "TransposeConvOptions",
                             "builtin_options": {
-                                "padding": "SAME",
+                                "padding": "VALID",
                                 "stride_w": )" + strideX + R"(,
                                 "stride_h": )" + strideY + R"(
                             },