diff options
author | Devansh Bansal/SNAP /SRI-Bangalore/Engineer/삼성전자 <b.devansh@samsung.com> | 2018-12-03 06:58:11 +0530 |
---|---|---|
committer | 오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 <hseok82.oh@samsung.com> | 2018-12-03 10:28:11 +0900 |
commit | 32be800d9a404c95f1cf6536fac6d5c4cb34c318 (patch) | |
tree | 57ece8f344dbcfee6e160a804b41d7b01e5f2f89 /include | |
parent | 903c05acc82bbeb2393c836d21a12a6ce801b718 (diff) | |
download | nnfw-32be800d9a404c95f1cf6536fac6d5c4cb34c318.tar.gz nnfw-32be800d9a404c95f1cf6536fac6d5c4cb34c318.tar.bz2 nnfw-32be800d9a404c95f1cf6536fac6d5c4cb34c318.zip |
Define TRANSPOSE_CONV_EX in NeuralNetworksEx.h (#3773)
This patch defines TRANSPOSE_CONV_EX in NeuralNetworksEx.h.
Related issue: #3347
Signed-off-by: b.devansh <b.devansh@samsung.com>
Diffstat (limited to 'include')
-rw-r--r-- | include/NeuralNetworksEx.h | 24 |
1 files changed, 24 insertions, 0 deletions
diff --git a/include/NeuralNetworksEx.h b/include/NeuralNetworksEx.h index d8c39e9e0..2c8fd9dc8 100644 --- a/include/NeuralNetworksEx.h +++ b/include/NeuralNetworksEx.h @@ -192,6 +192,30 @@ typedef enum { * * 0: A tensor of the same {@link OperandCode} as input0. */ ANEURALNETWORKS_TENSORFLOW_SUM_EX = 50009, + + /** + * A transposed convolutional layer carries out a regular convolution but reverts its spatisal transformation. + * Transpose convolution basically performs convolution with transposed weights. + * + * Supported tensor {@link OperandCode}: + * {@link ANEURALNETWORKS_TENSOR_FLOAT32} + * {@link ANEURALNETWORKS_TENSOR_INT32} + * + * Supported tensor rank: only 4 + * + * Inputs: + * 0: An {@link ANEURALNETWORKS_INT32} 1-D four element tensor, specifying the output shape. + * 1: A 4-D tensor, of shape [depth_out, filter_height, filter_width, depth_in], specifying the filter. + * 2: A 4-D tensor, of shape [batches, height, width, depth_in], specifying the input. + * 3: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding type. + * 4: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when + * walking through input in the ‘width’ dimension. + * 5: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when + * walking through input in the height dimension. + * + * Outputs: + * 0: The output 4-D tensor, of shape [batches, out_height, out_width, depth_out]. + */ ANEURALNETWORKS_TRANSPOSE_CONV_EX = 50010, /** |