Defined CUSTOM pack operation (#3648)
authorVishal Keshav/SNAP /SRI-Bangalore/Engineer/삼성전자 <vishal.k1@samsung.com>
Fri, 23 Nov 2018 02:38:40 +0000 (08:08 +0530)
committer오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 <hseok82.oh@samsung.com>
Fri, 23 Nov 2018 02:38:40 +0000 (11:38 +0900)
Added details for pack operation

Signed-off-by: Vishal Keshav <vishal.k1@samsung.com>
include/NeuralNetworksEx.h

index 45e6167..9ff8291 100644 (file)
@@ -228,6 +228,31 @@ typedef enum {
      */
     ANEURALNETWORKS_EQUAL_EX = 50011,
     ANEURALNETWORKS_ABS_EX = 50012,
+    /**
+     * Packs a list of rank-R tensors into one rank- (R+1) tensor along the axis dimension.
+     *
+     * The input tensors must have identical {@link OperandCode} and the same
+     * dimensions.
+     *
+     * Supported tensor {@link OperandCode}:
+     * * {@link ANEURALNETWORKS_TENSOR_FLOAT32}
+     * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}
+     *
+     * Supported tensor rank: up to 3
+     *
+     * Inputs:
+     * * 0 ~ n-1: The list of n input tensors, of shape
+     *            [D0, D1, ..., Daxis(i), ..., Dm]. For inputs of
+     *            {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, all input tensors
+     *            must have the same scale and zeroPoint.
+     * * n: An {@link ANEURALNETWORKS_INT32} scalar, specifying the
+     *      pack axis.
+     *
+     * Outputs:
+     * * 0: The output, a tensor of the same {@link OperandCode} as the input
+     *      tensors. The output shape is [D0, D1, ..., N at Daxis(i), ..., Dm+1]
+     *      where N is the number of tensors to be packed.
+     */
     ANEURALNETWORKS_PACK_EX = 50013,
     /**
      * Unpacks a given rank-R tensors into num_splits rank- (R-1) tensors along the axis dimention.