COMPMID-1367: Enable NHWC in graph examples
Change-Id: Iabc54a3a1bdcd46a9a921cda39c7c85fef672b72
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/141449
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Tested-by: Jenkins <bsgcomp@arm.com>
diff --git a/examples/graph_alexnet.cpp b/examples/graph_alexnet.cpp
index 63e7b16..944a435 100644
--- a/examples/graph_alexnet.cpp
+++ b/examples/graph_alexnet.cpp
@@ -60,7 +60,6 @@
// Checks
ARM_COMPUTE_EXIT_ON_MSG(arm_compute::is_data_type_quantized_asymmetric(common_params.data_type), "Unsupported data type!");
- ARM_COMPUTE_EXIT_ON_MSG(common_params.data_layout == DataLayout::NHWC, "Unsupported data layout!");
// Print parameter values
std::cout << common_params << std::endl;
@@ -72,14 +71,20 @@
const std::array<float, 3> mean_rgb{ { 122.68f, 116.67f, 104.01f } };
std::unique_ptr<IPreprocessor> preprocessor = arm_compute::support::cpp14::make_unique<CaffePreproccessor>(mean_rgb);
+ // Create input descriptor
+ const TensorShape tensor_shape = permute_shape(TensorShape(227U, 227U, 3U, 1U), DataLayout::NCHW, common_params.data_layout);
+ TensorDescriptor input_descriptor = TensorDescriptor(tensor_shape, common_params.data_type).set_layout(common_params.data_layout);
+
+ // Set weights trained layout
+ const DataLayout weights_layout = DataLayout::NCHW;
+
graph << common_params.target
<< common_params.fast_math_hint
- << InputLayer(TensorDescriptor(TensorShape(227U, 227U, 3U, 1U), common_params.data_type),
- get_input_accessor(common_params, std::move(preprocessor)))
+ << InputLayer(input_descriptor, get_input_accessor(common_params, std::move(preprocessor)))
// Layer 1
<< ConvolutionLayer(
11U, 11U, 96U,
- get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv1_b.npy"),
PadStrideInfo(4, 4, 0, 0))
.set_name("conv1")
@@ -89,7 +94,7 @@
// Layer 2
<< ConvolutionLayer(
5U, 5U, 256U,
- get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv2_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv2_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv2_b.npy"),
PadStrideInfo(1, 1, 2, 2), 2)
.set_name("conv2")
@@ -99,7 +104,7 @@
// Layer 3
<< ConvolutionLayer(
3U, 3U, 384U,
- get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv3_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv3_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv3_b.npy"),
PadStrideInfo(1, 1, 1, 1))
.set_name("conv3")
@@ -107,7 +112,7 @@
// Layer 4
<< ConvolutionLayer(
3U, 3U, 384U,
- get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv4_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv4_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv4_b.npy"),
PadStrideInfo(1, 1, 1, 1), 2)
.set_name("conv4")
@@ -115,7 +120,7 @@
// Layer 5
<< ConvolutionLayer(
3U, 3U, 256U,
- get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv5_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv5_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv5_b.npy"),
PadStrideInfo(1, 1, 1, 1), 2)
.set_name("conv5")
@@ -124,21 +129,21 @@
// Layer 6
<< FullyConnectedLayer(
4096U,
- get_weights_accessor(data_path, "/cnn_data/alexnet_model/fc6_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/alexnet_model/fc6_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/alexnet_model/fc6_b.npy"))
.set_name("fc6")
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("relu6")
// Layer 7
<< FullyConnectedLayer(
4096U,
- get_weights_accessor(data_path, "/cnn_data/alexnet_model/fc7_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/alexnet_model/fc7_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/alexnet_model/fc7_b.npy"))
.set_name("fc7")
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("relu7")
// Layer 8
<< FullyConnectedLayer(
1000U,
- get_weights_accessor(data_path, "/cnn_data/alexnet_model/fc8_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/alexnet_model/fc8_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/alexnet_model/fc8_b.npy"))
.set_name("fc8")
// Softmax
diff --git a/examples/graph_googlenet.cpp b/examples/graph_googlenet.cpp
index 4497dbd..d5bd0c0 100644
--- a/examples/graph_googlenet.cpp
+++ b/examples/graph_googlenet.cpp
@@ -60,7 +60,6 @@
// Checks
ARM_COMPUTE_EXIT_ON_MSG(arm_compute::is_data_type_quantized_asymmetric(common_params.data_type), "Unsupported data type!");
- ARM_COMPUTE_EXIT_ON_MSG(common_params.data_layout == DataLayout::NHWC, "Unsupported data layout!");
// Print parameter values
std::cout << common_params << std::endl;
@@ -72,13 +71,19 @@
const std::array<float, 3> mean_rgb{ { 122.68f, 116.67f, 104.01f } };
std::unique_ptr<IPreprocessor> preprocessor = arm_compute::support::cpp14::make_unique<CaffePreproccessor>(mean_rgb);
+ // Create input descriptor
+ const TensorShape tensor_shape = permute_shape(TensorShape(224U, 224U, 3U, 1U), DataLayout::NCHW, common_params.data_layout);
+ TensorDescriptor input_descriptor = TensorDescriptor(tensor_shape, common_params.data_type).set_layout(common_params.data_layout);
+
+ // Set weights trained layout
+ const DataLayout weights_layout = DataLayout::NCHW;
+
graph << common_params.target
<< common_params.fast_math_hint
- << InputLayer(TensorDescriptor(TensorShape(224U, 224U, 3U, 1U), common_params.data_type),
- get_input_accessor(common_params, std::move(preprocessor)))
+ << InputLayer(input_descriptor, get_input_accessor(common_params, std::move(preprocessor)))
<< ConvolutionLayer(
7U, 7U, 64U,
- get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv1/conv1_7x7_s2_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv1/conv1_7x7_s2_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv1/conv1_7x7_s2_b.npy"),
PadStrideInfo(2, 2, 3, 3))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
@@ -86,33 +91,33 @@
<< NormalizationLayer(NormalizationLayerInfo(NormType::CROSS_MAP, 5, 0.0001f, 0.75f))
<< ConvolutionLayer(
1U, 1U, 64U,
- get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_reduce_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_reduce_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_reduce_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(
3U, 3U, 192U,
- get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_b.npy"),
PadStrideInfo(1, 1, 1, 1))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< NormalizationLayer(NormalizationLayerInfo(NormType::CROSS_MAP, 5, 0.0001f, 0.75f))
<< PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)));
- graph << get_inception_node(data_path, "inception_3a", 64, std::make_tuple(96U, 128U), std::make_tuple(16U, 32U), 32U);
- graph << get_inception_node(data_path, "inception_3b", 128, std::make_tuple(128U, 192U), std::make_tuple(32U, 96U), 64U);
+ graph << get_inception_node(data_path, "inception_3a", weights_layout, 64, std::make_tuple(96U, 128U), std::make_tuple(16U, 32U), 32U);
+ graph << get_inception_node(data_path, "inception_3b", weights_layout, 128, std::make_tuple(128U, 192U), std::make_tuple(32U, 96U), 64U);
graph << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)));
- graph << get_inception_node(data_path, "inception_4a", 192, std::make_tuple(96U, 208U), std::make_tuple(16U, 48U), 64U);
- graph << get_inception_node(data_path, "inception_4b", 160, std::make_tuple(112U, 224U), std::make_tuple(24U, 64U), 64U);
- graph << get_inception_node(data_path, "inception_4c", 128, std::make_tuple(128U, 256U), std::make_tuple(24U, 64U), 64U);
- graph << get_inception_node(data_path, "inception_4d", 112, std::make_tuple(144U, 288U), std::make_tuple(32U, 64U), 64U);
- graph << get_inception_node(data_path, "inception_4e", 256, std::make_tuple(160U, 320U), std::make_tuple(32U, 128U), 128U);
+ graph << get_inception_node(data_path, "inception_4a", weights_layout, 192, std::make_tuple(96U, 208U), std::make_tuple(16U, 48U), 64U);
+ graph << get_inception_node(data_path, "inception_4b", weights_layout, 160, std::make_tuple(112U, 224U), std::make_tuple(24U, 64U), 64U);
+ graph << get_inception_node(data_path, "inception_4c", weights_layout, 128, std::make_tuple(128U, 256U), std::make_tuple(24U, 64U), 64U);
+ graph << get_inception_node(data_path, "inception_4d", weights_layout, 112, std::make_tuple(144U, 288U), std::make_tuple(32U, 64U), 64U);
+ graph << get_inception_node(data_path, "inception_4e", weights_layout, 256, std::make_tuple(160U, 320U), std::make_tuple(32U, 128U), 128U);
graph << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)));
- graph << get_inception_node(data_path, "inception_5a", 256, std::make_tuple(160U, 320U), std::make_tuple(32U, 128U), 128U);
- graph << get_inception_node(data_path, "inception_5b", 384, std::make_tuple(192U, 384U), std::make_tuple(48U, 128U), 128U);
+ graph << get_inception_node(data_path, "inception_5a", weights_layout, 256, std::make_tuple(160U, 320U), std::make_tuple(32U, 128U), 128U);
+ graph << get_inception_node(data_path, "inception_5b", weights_layout, 384, std::make_tuple(192U, 384U), std::make_tuple(48U, 128U), 128U);
graph << PoolingLayer(PoolingLayerInfo(PoolingType::AVG, 7, PadStrideInfo(1, 1, 0, 0, DimensionRoundingType::CEIL)))
<< FullyConnectedLayer(
1000U,
- get_weights_accessor(data_path, "/cnn_data/googlenet_model/loss3/loss3_classifier_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/googlenet_model/loss3/loss3_classifier_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/googlenet_model/loss3/loss3_classifier_b.npy"))
<< SoftmaxLayer()
<< OutputLayer(get_output_accessor(common_params, 5));
@@ -139,7 +144,7 @@
CommonGraphParams common_params;
Stream graph;
- BranchLayer get_inception_node(const std::string &data_path, std::string &¶m_path,
+ BranchLayer get_inception_node(const std::string &data_path, std::string &¶m_path, DataLayout weights_layout,
unsigned int a_filt,
std::tuple<unsigned int, unsigned int> b_filters,
std::tuple<unsigned int, unsigned int> c_filters,
@@ -149,7 +154,7 @@
SubStream i_a(graph);
i_a << ConvolutionLayer(
1U, 1U, a_filt,
- get_weights_accessor(data_path, total_path + "1x1_w.npy"),
+ get_weights_accessor(data_path, total_path + "1x1_w.npy", weights_layout),
get_weights_accessor(data_path, total_path + "1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
@@ -157,13 +162,13 @@
SubStream i_b(graph);
i_b << ConvolutionLayer(
1U, 1U, std::get<0>(b_filters),
- get_weights_accessor(data_path, total_path + "3x3_reduce_w.npy"),
+ get_weights_accessor(data_path, total_path + "3x3_reduce_w.npy", weights_layout),
get_weights_accessor(data_path, total_path + "3x3_reduce_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(
3U, 3U, std::get<1>(b_filters),
- get_weights_accessor(data_path, total_path + "3x3_w.npy"),
+ get_weights_accessor(data_path, total_path + "3x3_w.npy", weights_layout),
get_weights_accessor(data_path, total_path + "3x3_b.npy"),
PadStrideInfo(1, 1, 1, 1))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
@@ -171,13 +176,13 @@
SubStream i_c(graph);
i_c << ConvolutionLayer(
1U, 1U, std::get<0>(c_filters),
- get_weights_accessor(data_path, total_path + "5x5_reduce_w.npy"),
+ get_weights_accessor(data_path, total_path + "5x5_reduce_w.npy", weights_layout),
get_weights_accessor(data_path, total_path + "5x5_reduce_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(
5U, 5U, std::get<1>(c_filters),
- get_weights_accessor(data_path, total_path + "5x5_w.npy"),
+ get_weights_accessor(data_path, total_path + "5x5_w.npy", weights_layout),
get_weights_accessor(data_path, total_path + "5x5_b.npy"),
PadStrideInfo(1, 1, 2, 2))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
@@ -186,7 +191,7 @@
i_d << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL)))
<< ConvolutionLayer(
1U, 1U, d_filt,
- get_weights_accessor(data_path, total_path + "pool_proj_w.npy"),
+ get_weights_accessor(data_path, total_path + "pool_proj_w.npy", weights_layout),
get_weights_accessor(data_path, total_path + "pool_proj_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
diff --git a/examples/graph_inception_v3.cpp b/examples/graph_inception_v3.cpp
index 67f4e3c..c41b0a8 100644
--- a/examples/graph_inception_v3.cpp
+++ b/examples/graph_inception_v3.cpp
@@ -60,7 +60,6 @@
// Checks
ARM_COMPUTE_EXIT_ON_MSG(arm_compute::is_data_type_quantized_asymmetric(common_params.data_type), "Unsupported data type!");
- ARM_COMPUTE_EXIT_ON_MSG(common_params.data_layout == DataLayout::NHWC, "Unsupported data layout!");
// Print parameter values
std::cout << common_params << std::endl;
@@ -71,12 +70,18 @@
// Create a preprocessor object
std::unique_ptr<IPreprocessor> preprocessor = arm_compute::support::cpp14::make_unique<TFPreproccessor>();
+ // Create input descriptor
+ const TensorShape tensor_shape = permute_shape(TensorShape(299U, 299U, 3U, 1U), DataLayout::NCHW, common_params.data_layout);
+ TensorDescriptor input_descriptor = TensorDescriptor(tensor_shape, common_params.data_type).set_layout(common_params.data_layout);
+
+ // Set weights trained layout
+ const DataLayout weights_layout = DataLayout::NCHW;
+
graph << common_params.target
<< common_params.fast_math_hint
- << InputLayer(TensorDescriptor(TensorShape(299U, 299U, 3U, 1U), common_params.data_type),
- get_input_accessor(common_params, std::move(preprocessor), false))
+ << InputLayer(input_descriptor, get_input_accessor(common_params, std::move(preprocessor), false))
<< ConvolutionLayer(3U, 3U, 32U,
- get_weights_accessor(data_path, "/cnn_data/inceptionv3_model/Conv2d_1a_3x3_weights.npy"),
+ get_weights_accessor(data_path, "/cnn_data/inceptionv3_model/Conv2d_1a_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(2, 2, 0, 0))
.set_name("Conv2d_1a_3x3/convolution")
<< BatchNormalizationLayer(get_weights_accessor(data_path,
@@ -89,7 +94,7 @@
.set_name("Conv2d_1a_3x3/BatchNorm/batchnorm")
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("Conv2d_1a_3x3/Relu")
<< ConvolutionLayer(3U, 3U, 32U,
- get_weights_accessor(data_path, "/cnn_data/inceptionv3_model/Conv2d_2a_3x3_weights.npy"),
+ get_weights_accessor(data_path, "/cnn_data/inceptionv3_model/Conv2d_2a_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
.set_name("Conv2d_2a_3x3/convolution")
<< BatchNormalizationLayer(get_weights_accessor(data_path,
@@ -103,7 +108,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("Conv2d_2a_3x3/Relu")
<< ConvolutionLayer(3U, 3U, 64U,
- get_weights_accessor(data_path, "/cnn_data/inceptionv3_model/Conv2d_2b_3x3_weights.npy"),
+ get_weights_accessor(data_path, "/cnn_data/inceptionv3_model/Conv2d_2b_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 1, 1))
.set_name("Conv2d_2b_3x3/convolution")
<< BatchNormalizationLayer(get_weights_accessor(data_path,
@@ -119,7 +124,7 @@
<< PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))).set_name("MaxPool_3a_3x3/MaxPool")
<< ConvolutionLayer(1U, 1U, 80U,
- get_weights_accessor(data_path, "/cnn_data/inceptionv3_model/Conv2d_3b_1x1_weights.npy"),
+ get_weights_accessor(data_path, "/cnn_data/inceptionv3_model/Conv2d_3b_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
.set_name("Conv2d_3b_1x1/convolution")
<< BatchNormalizationLayer(get_weights_accessor(data_path,
@@ -133,7 +138,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("Conv2d_3b_1x1/Relu")
<< ConvolutionLayer(3U, 3U, 192U,
- get_weights_accessor(data_path, "/cnn_data/inceptionv3_model/Conv2d_4a_3x3_weights.npy"),
+ get_weights_accessor(data_path, "/cnn_data/inceptionv3_model/Conv2d_4a_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
.set_name("Conv2d_4a_3x3/convolution")
<< BatchNormalizationLayer(get_weights_accessor(data_path,
@@ -148,45 +153,45 @@
<< PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))).set_name("MaxPool_5a_3x3/MaxPool");
- graph << get_inception_node_A(data_path, "Mixed_5b", 64U, std::make_tuple(48U, 64U), std::make_tuple(64U, 96U, 96U),
+ graph << get_inception_node_A(data_path, "Mixed_5b", weights_layout, 64U, std::make_tuple(48U, 64U), std::make_tuple(64U, 96U, 96U),
32U)
.set_name("Mixed_5b/concat");
- graph << get_inception_node_A(data_path, "Mixed_5c", 64U, std::make_tuple(48U, 64U), std::make_tuple(64U, 96U, 96U),
+ graph << get_inception_node_A(data_path, "Mixed_5c", weights_layout, 64U, std::make_tuple(48U, 64U), std::make_tuple(64U, 96U, 96U),
64U, true)
.set_name("Mixed_5c/concat");
- graph << get_inception_node_A(data_path, "Mixed_5d", 64U, std::make_tuple(48U, 64U), std::make_tuple(64U, 96U, 96U),
+ graph << get_inception_node_A(data_path, "Mixed_5d", weights_layout, 64U, std::make_tuple(48U, 64U), std::make_tuple(64U, 96U, 96U),
64U)
.set_name("Mixed_5d/concat");
- graph << get_inception_node_B(data_path, "Mixed_6a", 384U, std::make_tuple(64U, 96U, 96U)).set_name("Mixed_6a/concat");
+ graph << get_inception_node_B(data_path, "Mixed_6a", weights_layout, 384U, std::make_tuple(64U, 96U, 96U)).set_name("Mixed_6a/concat");
- graph << get_inception_node_C(data_path, "Mixed_6b", 192U, std::make_tuple(128U, 128U, 192U),
+ graph << get_inception_node_C(data_path, "Mixed_6b", weights_layout, 192U, std::make_tuple(128U, 128U, 192U),
std::make_tuple(128U, 128U, 128U, 128U, 192U), 192U)
.set_name("Mixed_6b/concat");
- graph << get_inception_node_C(data_path, "Mixed_6c", 192U, std::make_tuple(160U, 160U, 192U),
+ graph << get_inception_node_C(data_path, "Mixed_6c", weights_layout, 192U, std::make_tuple(160U, 160U, 192U),
std::make_tuple(160U, 160U, 160U, 160U, 192U), 192U)
.set_name("Mixed_6c/concat");
- graph << get_inception_node_C(data_path, "Mixed_6d", 192U, std::make_tuple(160U, 160U, 192U),
+ graph << get_inception_node_C(data_path, "Mixed_6d", weights_layout, 192U, std::make_tuple(160U, 160U, 192U),
std::make_tuple(160U, 160U, 160U, 160U, 192U), 192U)
.set_name("Mixed_6d/concat");
- graph << get_inception_node_C(data_path, "Mixed_6e", 192U, std::make_tuple(192U, 192U, 192U),
+ graph << get_inception_node_C(data_path, "Mixed_6e", weights_layout, 192U, std::make_tuple(192U, 192U, 192U),
std::make_tuple(192U, 192U, 192U, 192U, 192U), 192U)
.set_name("Mixed_6e/concat");
- graph << get_inception_node_D(data_path, "Mixed_7a", std::make_tuple(192U, 320U),
+ graph << get_inception_node_D(data_path, "Mixed_7a", weights_layout, std::make_tuple(192U, 320U),
std::make_tuple(192U, 192U, 192U, 192U))
.set_name("Mixed_7a/concat");
- graph << get_inception_node_E(data_path, "Mixed_7b", 320U, std::make_tuple(384U, 384U, 384U),
+ graph << get_inception_node_E(data_path, "Mixed_7b", weights_layout, 320U, std::make_tuple(384U, 384U, 384U),
std::make_tuple(448U, 384U, 384U, 384U), 192U)
.set_name("Mixed_7b/concat");
- graph << get_inception_node_E(data_path, "Mixed_7c", 320U, std::make_tuple(384U, 384U, 384U),
+ graph << get_inception_node_E(data_path, "Mixed_7c", weights_layout, 320U, std::make_tuple(384U, 384U, 384U),
std::make_tuple(448U, 384U, 384U, 384U), 192U, true)
.set_name("Mixed_7c/concat");
graph << PoolingLayer(PoolingLayerInfo(PoolingType::AVG, 8, PadStrideInfo(1, 1, 0, 0, DimensionRoundingType::CEIL))).set_name("Logits/AvgPool_1a_8x8/AvgPool")
<< ConvolutionLayer(1U, 1U, 1001U, get_weights_accessor(data_path,
- "/cnn_data/inceptionv3_model/Logits_Conv2d_1c_1x1_weights.npy"),
+ "/cnn_data/inceptionv3_model/Logits_Conv2d_1c_1x1_weights.npy", weights_layout),
get_weights_accessor(data_path,
"/cnn_data/inceptionv3_model/Logits_Conv2d_1c_1x1_biases.npy"),
PadStrideInfo(1, 1, 0, 0))
@@ -218,7 +223,7 @@
Stream graph;
private:
- BranchLayer get_inception_node_A(const std::string &data_path, std::string &¶m_path,
+ BranchLayer get_inception_node_A(const std::string &data_path, std::string &¶m_path, DataLayout weights_layout,
unsigned int a_filt,
std::tuple<unsigned int, unsigned int> b_filters,
std::tuple<unsigned int, unsigned int, unsigned int> c_filters,
@@ -239,7 +244,7 @@
SubStream i_a(graph);
i_a << ConvolutionLayer(
1U, 1U, a_filt,
- get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(param_path + "/Branch_0/Conv2d_0a_1x1/convolution")
@@ -255,7 +260,7 @@
SubStream i_b(graph);
i_b << ConvolutionLayer(
1U, 1U, std::get<0>(b_filters),
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d" + conv_id0 + "1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d" + conv_id0 + "1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(param_path + "/Branch_1/Conv2d" + conv_id0 + "1x1/convolution")
@@ -269,7 +274,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/Branch_1/Conv2d" + conv_id0 + "1x1/Relu")
<< ConvolutionLayer(
5U, 5U, std::get<1>(b_filters),
- get_weights_accessor(data_path, total_path + "Branch_1_Conv" + conv_id1 + "5x5_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv" + conv_id1 + "5x5_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 2, 2))
.set_name(param_path + "/Branch_1/Conv2d" + conv_id1 + "5x5/convolution")
@@ -285,7 +290,7 @@
SubStream i_c(graph);
i_c << ConvolutionLayer(
1U, 1U, std::get<0>(c_filters),
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(param_path + "/Branch_2/Conv2d_0a_1x1/convolution")
@@ -299,7 +304,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/Branch_2/Conv2d_0a_1x1/Relu")
<< ConvolutionLayer(
3U, 3U, std::get<1>(c_filters),
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0b_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0b_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 1, 1))
.set_name(param_path + "/Branch_2/Conv2d_0b_3x3/convolution")
@@ -313,7 +318,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/Branch_2/Conv2d_0b_3x3/Relu")
<< ConvolutionLayer(
3U, 3U, std::get<2>(c_filters),
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0c_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0c_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 1, 1))
.set_name(param_path + "/Branch_2/Conv2d_0c_3x3/convolution")
@@ -330,7 +335,7 @@
i_d << PoolingLayer(PoolingLayerInfo(PoolingType::AVG, 3, PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL), true)).set_name(param_path + "/Branch_3/AvgPool_0a_3x3/AvgPool")
<< ConvolutionLayer(
1U, 1U, d_filt,
- get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(param_path + "/Branch_3/Conv2d_0b_1x1/convolution")
@@ -346,7 +351,7 @@
return BranchLayer(BranchMergeMethod::DEPTH_CONCATENATE, std::move(i_a), std::move(i_b), std::move(i_c), std::move(i_d));
}
- BranchLayer get_inception_node_B(const std::string &data_path, std::string &¶m_path,
+ BranchLayer get_inception_node_B(const std::string &data_path, std::string &¶m_path, DataLayout weights_layout,
unsigned int a_filt,
std::tuple<unsigned int, unsigned int, unsigned int> b_filters)
{
@@ -354,7 +359,7 @@
SubStream i_a(graph);
i_a << ConvolutionLayer(
3U, 3U, a_filt,
- get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(2, 2, 0, 0))
.set_name(param_path + "/Branch_0/Conv2d_1a_1x1/convolution")
@@ -370,7 +375,7 @@
SubStream i_b(graph);
i_b << ConvolutionLayer(
1U, 1U, std::get<0>(b_filters),
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(param_path + "/Branch_1/Conv2d_0a_1x1/convolution")
@@ -384,7 +389,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/Branch_1/Conv2d_0a_1x1/Relu")
<< ConvolutionLayer(
3U, 3U, std::get<1>(b_filters),
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 1, 1))
.set_name(param_path + "/Branch_1/Conv2d_0b_3x3/convolution")
@@ -398,7 +403,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/Branch_1/Conv2d_0b_3x3/Relu")
<< ConvolutionLayer(
3U, 3U, std::get<2>(b_filters),
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_1a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_1a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(2, 2, 0, 0))
.set_name(param_path + "/Branch_1/Conv2d_1a_1x1/convolution")
@@ -417,7 +422,7 @@
return BranchLayer(BranchMergeMethod::DEPTH_CONCATENATE, std::move(i_a), std::move(i_b), std::move(i_c));
}
- BranchLayer get_inception_node_C(const std::string &data_path, std::string &¶m_path,
+ BranchLayer get_inception_node_C(const std::string &data_path, std::string &¶m_path, DataLayout weights_layout,
unsigned int a_filt,
std::tuple<unsigned int, unsigned int, unsigned int> b_filters,
std::tuple<unsigned int, unsigned int, unsigned int, unsigned int, unsigned int> c_filters,
@@ -427,7 +432,7 @@
SubStream i_a(graph);
i_a << ConvolutionLayer(
1U, 1U, a_filt,
- get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(param_path + "/Branch_0/Conv2d_0a_1x1/convolution")
@@ -443,7 +448,7 @@
SubStream i_b(graph);
i_b << ConvolutionLayer(
1U, 1U, std::get<0>(b_filters),
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(param_path + "/Branch_1/Conv2d_0a_1x1/convolution")
@@ -457,7 +462,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/Branch_1/Conv2d_0a_1x1/Relu")
<< ConvolutionLayer(
7U, 1U, std::get<1>(b_filters),
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x7_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x7_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 3, 0))
.set_name(param_path + "/Branch_1/Conv2d_0b_1x7/convolution")
@@ -471,7 +476,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/Branch_1/Conv2d_0b_1x7/Relu")
<< ConvolutionLayer(
1U, 7U, std::get<2>(b_filters),
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_7x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_7x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 3))
.set_name(param_path + "/Branch_1/Conv2d_0c_7x1/convolution")
@@ -487,7 +492,7 @@
SubStream i_c(graph);
i_c << ConvolutionLayer(
1U, 1U, std::get<0>(c_filters),
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(param_path + "/Branch_2/Conv2d_0a_1x1/convolution")
@@ -501,7 +506,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/Branch_2/Conv2d_0a_1x1/Relu")
<< ConvolutionLayer(
1U, 7U, std::get<1>(c_filters),
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0b_7x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0b_7x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 3))
.set_name(param_path + "/Branch_2/Conv2d_0b_7x1/convolution")
@@ -515,7 +520,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/Branch_2/Conv2d_0b_7x1/Relu")
<< ConvolutionLayer(
7U, 1U, std::get<2>(c_filters),
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0c_1x7_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0c_1x7_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 3, 0))
.set_name(param_path + "/Branch_2/Conv2d_0c_1x7/convolution")
@@ -529,7 +534,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/Branch_2/Conv2d_0c_1x7/Relu")
<< ConvolutionLayer(
1U, 7U, std::get<3>(c_filters),
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0d_7x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0d_7x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 3))
.set_name(param_path + "/Branch_2/Conv2d_0d_7x1/convolution")
@@ -543,7 +548,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/Branch_2/Conv2d_0d_7x1/Relu")
<< ConvolutionLayer(
7U, 1U, std::get<4>(c_filters),
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0e_1x7_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0e_1x7_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 3, 0))
.set_name(param_path + "/Branch_2/Conv2d_0e_1x7/convolution")
@@ -560,7 +565,7 @@
i_d << PoolingLayer(PoolingLayerInfo(PoolingType::AVG, 3, PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL), true)).set_name(param_path + "/Branch_3/AvgPool_0a_3x3/AvgPool")
<< ConvolutionLayer(
1U, 1U, d_filt,
- get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(param_path + "/Branch_3/Conv2d_0b_1x1/convolution")
@@ -576,15 +581,15 @@
return BranchLayer(BranchMergeMethod::DEPTH_CONCATENATE, std::move(i_a), std::move(i_b), std::move(i_c), std::move(i_d));
}
- BranchLayer get_inception_node_D(const std::string &data_path, std::string &¶m_path,
- std::tuple<unsigned int, unsigned int> a_filters,
+ BranchLayer get_inception_node_D(const std::string &data_path, std::string &¶m_path, DataLayout weights_layout,
+ std::tuple<unsigned int, unsigned int> a_filters,
std::tuple<unsigned int, unsigned int, unsigned int, unsigned int> b_filters)
{
std::string total_path = "/cnn_data/inceptionv3_model/" + param_path + "_";
SubStream i_a(graph);
i_a << ConvolutionLayer(
1U, 1U, std::get<0>(a_filters),
- get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(param_path + "/Branch_0/Conv2d_0a_1x1/convolution")
@@ -598,7 +603,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/Branch_0/Conv2d_0a_1x1/Relu")
<< ConvolutionLayer(
3U, 3U, std::get<1>(a_filters),
- get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(2, 2, 0, 0))
.set_name(param_path + "/Branch_0/Conv2d_1a_3x3/convolution")
@@ -614,7 +619,7 @@
SubStream i_b(graph);
i_b << ConvolutionLayer(
1U, 1U, std::get<0>(b_filters),
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(param_path + "/Branch_1/Conv2d_0a_1x1/convolution")
@@ -628,7 +633,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/Branch_1/Conv2d_0a_1x1/Relu")
<< ConvolutionLayer(
7U, 1U, std::get<1>(b_filters),
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x7_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x7_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 3, 0))
.set_name(param_path + "/Branch_1/Conv2d_0b_1x7/convolution")
@@ -642,7 +647,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/Branch_1/Conv2d_0b_1x7/Relu")
<< ConvolutionLayer(
1U, 7U, std::get<2>(b_filters),
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_7x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_7x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 3))
.set_name(param_path + "/Branch_1/Conv2d_0c_7x1/convolution")
@@ -656,7 +661,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/Branch_1/Conv2d_0c_7x1/Relu")
<< ConvolutionLayer(
3U, 3U, std::get<3>(b_filters),
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_1a_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_1a_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(2, 2, 0, 0))
.set_name(param_path + "/Branch_1/Conv2d_1a_3x3/convolution")
@@ -675,7 +680,7 @@
return BranchLayer(BranchMergeMethod::DEPTH_CONCATENATE, std::move(i_a), std::move(i_b), std::move(i_c));
}
- BranchLayer get_inception_node_E(const std::string &data_path, std::string &¶m_path,
+ BranchLayer get_inception_node_E(const std::string &data_path, std::string &¶m_path, DataLayout weights_layout,
unsigned int a_filt,
std::tuple<unsigned int, unsigned int, unsigned int> b_filters,
std::tuple<unsigned int, unsigned int, unsigned int, unsigned int> c_filters,
@@ -693,7 +698,7 @@
SubStream i_a(graph);
i_a << ConvolutionLayer(
1U, 1U, a_filt,
- get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(param_path + "/Branch_0/Conv2d_0a_1x1/convolution")
@@ -709,7 +714,7 @@
SubStream i_b(graph);
i_b << ConvolutionLayer(
1U, 1U, std::get<0>(b_filters),
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(param_path + "/Branch_1/Conv2d_0a_1x1/convolution")
@@ -725,7 +730,7 @@
SubStream i_b1(i_b);
i_b1 << ConvolutionLayer(
3U, 1U, std::get<1>(b_filters),
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 1, 0))
.set_name(param_path + "/Branch_1/Conv2d_0b_1x3/convolution")
@@ -741,7 +746,7 @@
SubStream i_b2(i_b);
i_b2 << ConvolutionLayer(
1U, 3U, std::get<2>(b_filters),
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d" + conv_id + "3x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d" + conv_id + "3x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 1))
.set_name(param_path + "/Branch_1/Conv2d" + conv_id + "3x1/convolution")
@@ -760,7 +765,7 @@
SubStream i_c(graph);
i_c << ConvolutionLayer(
1U, 1U, std::get<0>(c_filters),
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(param_path + "/Branch_2/Conv2d_0a_1x1/convolution")
@@ -774,7 +779,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/Branch_2/Conv2d_0a_1x1/Relu")
<< ConvolutionLayer(
3U, 3U, std::get<1>(c_filters),
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0b_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0b_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 1, 1))
.set_name(param_path + "/Branch_2/Conv2d_0b_3x3/convolution")
@@ -790,7 +795,7 @@
SubStream i_c1(i_c);
i_c1 << ConvolutionLayer(
3U, 1U, std::get<2>(c_filters),
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0c_1x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0c_1x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 1, 0))
.set_name(param_path + "/Branch_2/Conv2d_0c_1x3/convolution")
@@ -806,7 +811,7 @@
SubStream i_c2(i_c);
i_c2 << ConvolutionLayer(
1U, 3U, std::get<3>(c_filters),
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0d_3x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0d_3x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 1))
.set_name(param_path + "/Branch_2/Conv2d_0d_3x1/convolution")
@@ -826,7 +831,7 @@
i_d << PoolingLayer(PoolingLayerInfo(PoolingType::AVG, 3, PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL), true)).set_name(param_path + "/Branch_3/AvgPool_0a_3x3/AvgPool")
<< ConvolutionLayer(
1U, 1U, d_filt,
- get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(param_path + "/Branch_3/Conv2d_0b_1x1/convolution")
diff --git a/examples/graph_inception_v4.cpp b/examples/graph_inception_v4.cpp
index 4e40592..b61acfc 100644
--- a/examples/graph_inception_v4.cpp
+++ b/examples/graph_inception_v4.cpp
@@ -60,7 +60,6 @@
// Checks
ARM_COMPUTE_EXIT_ON_MSG(arm_compute::is_data_type_quantized_asymmetric(common_params.data_type), "Unsupported data type!");
- ARM_COMPUTE_EXIT_ON_MSG(common_params.data_layout == DataLayout::NHWC, "Unsupported data layout!");
// Print parameter values
std::cout << common_params << std::endl;
@@ -71,13 +70,19 @@
// Create a preprocessor object
std::unique_ptr<IPreprocessor> preprocessor = arm_compute::support::cpp14::make_unique<TFPreproccessor>();
+ // Create input descriptor
+ const TensorShape tensor_shape = permute_shape(TensorShape(299U, 299U, 3U, 1U), DataLayout::NCHW, common_params.data_layout);
+ TensorDescriptor input_descriptor = TensorDescriptor(tensor_shape, common_params.data_type).set_layout(common_params.data_layout);
+
+ // Set weights trained layout
+ const DataLayout weights_layout = DataLayout::NCHW;
+
graph << common_params.target
<< common_params.fast_math_hint
- << InputLayer(TensorDescriptor(TensorShape(299U, 299U, 3U, 1U), common_params.data_type),
- get_input_accessor(common_params, std::move(preprocessor), false))
+ << InputLayer(input_descriptor, get_input_accessor(common_params, std::move(preprocessor), false))
// Conv2d_1a_3x3
<< ConvolutionLayer(3U, 3U, 32U,
- get_weights_accessor(data_path, "/cnn_data/inceptionv4_model/Conv2d_1a_3x3_weights.npy"),
+ get_weights_accessor(data_path, "/cnn_data/inceptionv4_model/Conv2d_1a_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(2, 2, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, "/cnn_data/inceptionv4_model/Conv2d_1a_3x3_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, "/cnn_data/inceptionv4_model/Conv2d_1a_3x3_BatchNorm_moving_variance.npy"),
@@ -87,7 +92,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
// Conv2d_2a_3x3
<< ConvolutionLayer(3U, 3U, 32U,
- get_weights_accessor(data_path, "/cnn_data/inceptionv4_model/Conv2d_2a_3x3_weights.npy"),
+ get_weights_accessor(data_path, "/cnn_data/inceptionv4_model/Conv2d_2a_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, "/cnn_data/inceptionv4_model/Conv2d_2a_3x3_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, "/cnn_data/inceptionv4_model/Conv2d_2a_3x3_BatchNorm_moving_variance.npy"),
@@ -97,7 +102,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
// Conv2d_2b_3x3
<< ConvolutionLayer(3U, 3U, 64U,
- get_weights_accessor(data_path, "/cnn_data/inceptionv4_model/Conv2d_2b_3x3_weights.npy"),
+ get_weights_accessor(data_path, "/cnn_data/inceptionv4_model/Conv2d_2b_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 1, 1))
<< BatchNormalizationLayer(get_weights_accessor(data_path, "/cnn_data/inceptionv4_model/Conv2d_2b_3x3_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, "/cnn_data/inceptionv4_model/Conv2d_2b_3x3_BatchNorm_moving_variance.npy"),
@@ -106,35 +111,35 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- graph << get_mixed_3a(data_path);
- graph << get_mixed_4a(data_path);
- graph << get_mixed_5a(data_path);
+ graph << get_mixed_3a(data_path, weights_layout);
+ graph << get_mixed_4a(data_path, weights_layout);
+ graph << get_mixed_5a(data_path, weights_layout);
// 4 inception A blocks
- graph << get_inceptionA_block(data_path, "Mixed_5b");
- graph << get_inceptionA_block(data_path, "Mixed_5c");
- graph << get_inceptionA_block(data_path, "Mixed_5d");
- graph << get_inceptionA_block(data_path, "Mixed_5e");
+ graph << get_inceptionA_block(data_path, weights_layout, "Mixed_5b");
+ graph << get_inceptionA_block(data_path, weights_layout, "Mixed_5c");
+ graph << get_inceptionA_block(data_path, weights_layout, "Mixed_5d");
+ graph << get_inceptionA_block(data_path, weights_layout, "Mixed_5e");
// reduction A block
- graph << get_reductionA_block(data_path);
+ graph << get_reductionA_block(data_path, weights_layout);
// 7 inception B blocks
- graph << get_inceptionB_block(data_path, "Mixed_6b");
- graph << get_inceptionB_block(data_path, "Mixed_6c");
- graph << get_inceptionB_block(data_path, "Mixed_6d");
- graph << get_inceptionB_block(data_path, "Mixed_6e");
- graph << get_inceptionB_block(data_path, "Mixed_6f");
- graph << get_inceptionB_block(data_path, "Mixed_6g");
- graph << get_inceptionB_block(data_path, "Mixed_6h");
+ graph << get_inceptionB_block(data_path, weights_layout, "Mixed_6b");
+ graph << get_inceptionB_block(data_path, weights_layout, "Mixed_6c");
+ graph << get_inceptionB_block(data_path, weights_layout, "Mixed_6d");
+ graph << get_inceptionB_block(data_path, weights_layout, "Mixed_6e");
+ graph << get_inceptionB_block(data_path, weights_layout, "Mixed_6f");
+ graph << get_inceptionB_block(data_path, weights_layout, "Mixed_6g");
+ graph << get_inceptionB_block(data_path, weights_layout, "Mixed_6h");
// reduction B block
- graph << get_reductionB_block(data_path);
+ graph << get_reductionB_block(data_path, weights_layout);
// 3 inception C blocks
- graph << get_inceptionC_block(data_path, "Mixed_7b");
- graph << get_inceptionC_block(data_path, "Mixed_7c");
- graph << get_inceptionC_block(data_path, "Mixed_7d");
+ graph << get_inceptionC_block(data_path, weights_layout, "Mixed_7b");
+ graph << get_inceptionC_block(data_path, weights_layout, "Mixed_7c");
+ graph << get_inceptionC_block(data_path, weights_layout, "Mixed_7d");
graph << PoolingLayer(PoolingLayerInfo(PoolingType::AVG))
<< FlattenLayer()
<< FullyConnectedLayer(
1001U,
- get_weights_accessor(data_path, "/cnn_data/inceptionv4_model/Logits_Logits_weights.npy"),
+ get_weights_accessor(data_path, "/cnn_data/inceptionv4_model/Logits_Logits_weights.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/inceptionv4_model/Logits_Logits_biases.npy"))
<< SoftmaxLayer()
<< OutputLayer(get_output_accessor(common_params, 5));
@@ -162,7 +167,7 @@
Stream graph;
private:
- BranchLayer get_mixed_3a(const std::string &data_path)
+ BranchLayer get_mixed_3a(const std::string &data_path, DataLayout weights_layout)
{
std::string total_path = "/cnn_data/inceptionv4_model/Mixed_3a_";
@@ -171,7 +176,7 @@
SubStream i_b(graph);
i_b << ConvolutionLayer(3U, 3U, 96U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(2, 2, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_3x3_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_3x3_BatchNorm_moving_variance.npy"),
@@ -183,13 +188,13 @@
return BranchLayer(BranchMergeMethod::DEPTH_CONCATENATE, std::move(i_a), std::move(i_b));
}
- BranchLayer get_mixed_4a(const std::string &data_path)
+ BranchLayer get_mixed_4a(const std::string &data_path, DataLayout weights_layout)
{
std::string total_path = "/cnn_data/inceptionv4_model/Mixed_4a_";
SubStream i_a(graph);
i_a << ConvolutionLayer(1U, 1U, 64U,
- get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_BatchNorm_moving_variance.npy"),
@@ -198,7 +203,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(3U, 3U, 96U,
- get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_BatchNorm_moving_variance.npy"),
@@ -209,7 +214,7 @@
SubStream i_b(graph);
i_b << ConvolutionLayer(1U, 1U, 64U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_BatchNorm_moving_variance.npy"),
@@ -218,7 +223,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(7U, 1U, 64U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x7_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x7_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 3, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x7_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x7_BatchNorm_moving_variance.npy"),
@@ -227,7 +232,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(1U, 7U, 64U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_7x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_7x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 3))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_7x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_7x1_BatchNorm_moving_variance.npy"),
@@ -236,7 +241,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(3U, 3U, 96U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_1a_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_1a_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_1a_3x3_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_1a_3x3_BatchNorm_moving_variance.npy"),
@@ -248,13 +253,13 @@
return BranchLayer(BranchMergeMethod::DEPTH_CONCATENATE, std::move(i_a), std::move(i_b));
}
- BranchLayer get_mixed_5a(const std::string &data_path)
+ BranchLayer get_mixed_5a(const std::string &data_path, DataLayout weights_layout)
{
std::string total_path = "/cnn_data/inceptionv4_model/Mixed_5a_";
SubStream i_a(graph);
i_a << ConvolutionLayer(3U, 3U, 192U,
- get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(2, 2, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_BatchNorm_moving_variance.npy"),
@@ -269,13 +274,13 @@
return BranchLayer(BranchMergeMethod::DEPTH_CONCATENATE, std::move(i_a), std::move(i_b));
}
- BranchLayer get_inceptionA_block(const std::string &data_path, std::string &¶m_path)
+ BranchLayer get_inceptionA_block(const std::string &data_path, DataLayout weights_layout, std::string &¶m_path)
{
std::string total_path = "/cnn_data/inceptionv4_model/" + param_path + "_";
SubStream i_a(graph);
i_a << ConvolutionLayer(1U, 1U, 96U,
- get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_BatchNorm_moving_variance.npy"),
@@ -286,7 +291,7 @@
SubStream i_b(graph);
i_b << ConvolutionLayer(1U, 1U, 64U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_BatchNorm_moving_variance.npy"),
@@ -295,7 +300,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(3U, 3U, 96U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 1, 1))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_3x3_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_3x3_BatchNorm_moving_variance.npy"),
@@ -306,7 +311,7 @@
SubStream i_c(graph);
i_c << ConvolutionLayer(1U, 1U, 64U,
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0a_1x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0a_1x1_BatchNorm_moving_variance.npy"),
@@ -315,7 +320,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(3U, 3U, 96U,
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0b_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0b_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 1, 1))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0b_3x3_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0b_3x3_BatchNorm_moving_variance.npy"),
@@ -324,7 +329,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(3U, 3U, 96U,
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0c_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0c_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 1, 1))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0c_3x3_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0c_3x3_BatchNorm_moving_variance.npy"),
@@ -336,7 +341,7 @@
SubStream i_d(graph);
i_d << PoolingLayer(PoolingLayerInfo(PoolingType::AVG, 3, PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL), true))
<< ConvolutionLayer(1U, 1U, 96U,
- get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_BatchNorm_moving_variance.npy"),
@@ -348,13 +353,13 @@
return BranchLayer(BranchMergeMethod::DEPTH_CONCATENATE, std::move(i_a), std::move(i_b), std::move(i_c), std::move(i_d));
}
- BranchLayer get_reductionA_block(const std::string &data_path)
+ BranchLayer get_reductionA_block(const std::string &data_path, DataLayout weights_layout)
{
std::string total_path = "/cnn_data/inceptionv4_model/Mixed_6a_";
SubStream i_a(graph);
i_a << ConvolutionLayer(3U, 3U, 384U,
- get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(2, 2, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_BatchNorm_moving_variance.npy"),
@@ -365,7 +370,7 @@
SubStream i_b(graph);
i_b << ConvolutionLayer(1U, 1U, 192U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_BatchNorm_moving_variance.npy"),
@@ -374,7 +379,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(3U, 3U, 224U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 1, 1))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_3x3_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_3x3_BatchNorm_moving_variance.npy"),
@@ -383,7 +388,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(3U, 3U, 256U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_1a_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_1a_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(2, 2, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_1a_3x3_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_1a_3x3_BatchNorm_moving_variance.npy"),
@@ -398,13 +403,13 @@
return BranchLayer(BranchMergeMethod::DEPTH_CONCATENATE, std::move(i_a), std::move(i_b), std::move(i_c));
}
- BranchLayer get_inceptionB_block(const std::string &data_path, std::string &¶m_path)
+ BranchLayer get_inceptionB_block(const std::string &data_path, DataLayout weights_layout, std::string &¶m_path)
{
std::string total_path = "/cnn_data/inceptionv4_model/" + param_path + "_";
SubStream i_a(graph);
i_a << ConvolutionLayer(1U, 1U, 384U,
- get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_BatchNorm_moving_variance.npy"),
@@ -415,7 +420,7 @@
SubStream i_b(graph);
i_b << ConvolutionLayer(1U, 1U, 192U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_BatchNorm_moving_variance.npy"),
@@ -424,7 +429,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(7U, 1U, 224U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x7_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x7_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 3, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x7_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x7_BatchNorm_moving_variance.npy"),
@@ -433,7 +438,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(1U, 7U, 256U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_7x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_7x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 3))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_7x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_7x1_BatchNorm_moving_variance.npy"),
@@ -444,7 +449,7 @@
SubStream i_c(graph);
i_c << ConvolutionLayer(1U, 1U, 192U,
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0a_1x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0a_1x1_BatchNorm_moving_variance.npy"),
@@ -453,7 +458,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(1U, 7U, 192U,
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0b_7x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0b_7x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 3))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0b_7x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0b_7x1_BatchNorm_moving_variance.npy"),
@@ -462,7 +467,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(7U, 1U, 224U,
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0c_1x7_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0c_1x7_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 3, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0c_1x7_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0c_1x7_BatchNorm_moving_variance.npy"),
@@ -471,7 +476,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(1U, 7U, 224U,
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0d_7x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0d_7x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 3))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0d_7x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0d_7x1_BatchNorm_moving_variance.npy"),
@@ -480,7 +485,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(7U, 1U, 256U,
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0e_1x7_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0e_1x7_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 3, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0e_1x7_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0e_1x7_BatchNorm_moving_variance.npy"),
@@ -492,7 +497,7 @@
SubStream i_d(graph);
i_d << PoolingLayer(PoolingLayerInfo(PoolingType::AVG, 3, PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL), true))
<< ConvolutionLayer(1U, 1U, 128U,
- get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_BatchNorm_moving_variance.npy"),
@@ -504,13 +509,13 @@
return BranchLayer(BranchMergeMethod::DEPTH_CONCATENATE, std::move(i_a), std::move(i_b), std::move(i_c), std::move(i_d));
}
- BranchLayer get_reductionB_block(const std::string &data_path)
+ BranchLayer get_reductionB_block(const std::string &data_path, DataLayout weights_layout)
{
std::string total_path = "/cnn_data/inceptionv4_model/Mixed_7a_";
SubStream i_a(graph);
i_a << ConvolutionLayer(1U, 1U, 192U,
- get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_BatchNorm_moving_variance.npy"),
@@ -519,7 +524,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(3U, 3U, 192U,
- get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(2, 2, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_1a_3x3_BatchNorm_moving_variance.npy"),
@@ -530,7 +535,7 @@
SubStream i_b(graph);
i_b << ConvolutionLayer(1U, 1U, 256U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_BatchNorm_moving_variance.npy"),
@@ -539,7 +544,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(7U, 1U, 256U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x7_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x7_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 3, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x7_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x7_BatchNorm_moving_variance.npy"),
@@ -548,7 +553,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(1U, 7U, 320U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_7x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_7x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 3))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_7x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_7x1_BatchNorm_moving_variance.npy"),
@@ -557,7 +562,7 @@
0.001f)
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(3U, 3U, 320U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_1a_3x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_1a_3x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(2, 2, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_1a_3x3_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_1a_3x3_BatchNorm_moving_variance.npy"),
@@ -572,13 +577,13 @@
return BranchLayer(BranchMergeMethod::DEPTH_CONCATENATE, std::move(i_a), std::move(i_b), std::move(i_c));
}
- BranchLayer get_inceptionC_block(const std::string &data_path, std::string &¶m_path)
+ BranchLayer get_inceptionC_block(const std::string &data_path, DataLayout weights_layout, std::string &¶m_path)
{
std::string total_path = "/cnn_data/inceptionv4_model/" + param_path + "_";
SubStream i_a(graph);
i_a << ConvolutionLayer(1U, 1U, 256U,
- get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_0_Conv2d_0a_1x1_BatchNorm_moving_variance.npy"),
@@ -590,7 +595,7 @@
SubStream i_b(graph);
i_b << ConvolutionLayer(
1U, 1U, 384U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(
@@ -604,7 +609,7 @@
SubStream i_b1(i_b);
i_b1 << ConvolutionLayer(
3U, 1U, 256U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0b_1x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 1, 0))
<< BatchNormalizationLayer(
@@ -618,7 +623,7 @@
SubStream i_b2(i_b);
i_b2 << ConvolutionLayer(
1U, 3U, 256U,
- get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_3x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_1_Conv2d_0c_3x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 1))
<< BatchNormalizationLayer(
@@ -635,7 +640,7 @@
SubStream i_c(graph);
i_c << ConvolutionLayer(
1U, 1U, 384U,
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0a_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0a_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(
@@ -647,7 +652,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(
1U, 3U, 448U,
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0b_3x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0b_3x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 1))
<< BatchNormalizationLayer(
@@ -659,7 +664,7 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< ConvolutionLayer(
3U, 1U, 512U,
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0c_1x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0c_1x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 1, 0))
<< BatchNormalizationLayer(
@@ -673,7 +678,7 @@
SubStream i_c1(i_c);
i_c1 << ConvolutionLayer(
3U, 1U, 256U,
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0d_1x3_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0d_1x3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 1, 0))
<< BatchNormalizationLayer(
@@ -687,7 +692,7 @@
SubStream i_c2(i_c);
i_c2 << ConvolutionLayer(
1U, 3U, 256U,
- get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0e_3x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_2_Conv2d_0e_3x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 1))
<< BatchNormalizationLayer(
@@ -704,7 +709,7 @@
SubStream i_d(graph);
i_d << PoolingLayer(PoolingLayerInfo(PoolingType::AVG, 3, PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL), true))
<< ConvolutionLayer(1U, 1U, 256U,
- get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_weights.npy"),
+ get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), PadStrideInfo(1, 1, 0, 0))
<< BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_BatchNorm_moving_mean.npy"),
get_weights_accessor(data_path, total_path + "Branch_3_Conv2d_0b_1x1_BatchNorm_moving_variance.npy"),
diff --git a/examples/graph_mobilenet.cpp b/examples/graph_mobilenet.cpp
index a747b3c..d182844 100644
--- a/examples/graph_mobilenet.cpp
+++ b/examples/graph_mobilenet.cpp
@@ -85,7 +85,7 @@
// Set graph hints
graph << common_params.target
- << DepthwiseConvolutionMethod::OPTIMIZED_3x3 // FIXME(COMPMID-1073): Add heuristics to automatically call the optimized 3x3 method
+ << DepthwiseConvolutionMethod::Optimized3x3 // FIXME(COMPMID-1073): Add heuristics to automatically call the optimized 3x3 method
<< common_params.fast_math_hint;
// Create core graph
diff --git a/examples/graph_resnet50.cpp b/examples/graph_resnet50.cpp
index 58f36f6..0ad719a 100644
--- a/examples/graph_resnet50.cpp
+++ b/examples/graph_resnet50.cpp
@@ -60,7 +60,6 @@
// Checks
ARM_COMPUTE_EXIT_ON_MSG(arm_compute::is_data_type_quantized_asymmetric(common_params.data_type), "Unsupported data type!");
- ARM_COMPUTE_EXIT_ON_MSG(common_params.data_layout == DataLayout::NHWC, "Unsupported data layout!");
// Print parameter values
std::cout << common_params << std::endl;
@@ -72,13 +71,20 @@
const std::array<float, 3> mean_rgb{ { 122.68f, 116.67f, 104.01f } };
std::unique_ptr<IPreprocessor> preprocessor = arm_compute::support::cpp14::make_unique<CaffePreproccessor>(mean_rgb,
false /* Do not convert to BGR */);
+
+ // Create input descriptor
+ const TensorShape tensor_shape = permute_shape(TensorShape(224U, 224U, 3U, 1U), DataLayout::NCHW, common_params.data_layout);
+ TensorDescriptor input_descriptor = TensorDescriptor(tensor_shape, common_params.data_type).set_layout(common_params.data_layout);
+
+ // Set weights trained layout
+ const DataLayout weights_layout = DataLayout::NCHW;
+
graph << common_params.target
<< common_params.fast_math_hint
- << InputLayer(TensorDescriptor(TensorShape(224U, 224U, 3U, 1U), common_params.data_type),
- get_input_accessor(common_params, std::move(preprocessor), false /* Do not convert to BGR */))
+ << InputLayer(input_descriptor, get_input_accessor(common_params, std::move(preprocessor), false /* Do not convert to BGR */))
<< ConvolutionLayer(
7U, 7U, 64U,
- get_weights_accessor(data_path, "/cnn_data/resnet50_model/conv1_weights.npy"),
+ get_weights_accessor(data_path, "/cnn_data/resnet50_model/conv1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(2, 2, 3, 3))
.set_name("conv1/convolution")
@@ -92,15 +98,15 @@
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv1/Relu")
<< PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 1, 0, 1, DimensionRoundingType::FLOOR))).set_name("pool1/MaxPool");
- add_residual_block(data_path, "block1", 64, 3, 2);
- add_residual_block(data_path, "block2", 128, 4, 2);
- add_residual_block(data_path, "block3", 256, 6, 2);
- add_residual_block(data_path, "block4", 512, 3, 1);
+ add_residual_block(data_path, "block1", weights_layout, 64, 3, 2);
+ add_residual_block(data_path, "block2", weights_layout, 128, 4, 2);
+ add_residual_block(data_path, "block3", weights_layout, 256, 6, 2);
+ add_residual_block(data_path, "block4", weights_layout, 512, 3, 1);
graph << PoolingLayer(PoolingLayerInfo(PoolingType::AVG)).set_name("pool5")
<< ConvolutionLayer(
1U, 1U, 1000U,
- get_weights_accessor(data_path, "/cnn_data/resnet50_model/logits_weights.npy"),
+ get_weights_accessor(data_path, "/cnn_data/resnet50_model/logits_weights.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/resnet50_model/logits_biases.npy"),
PadStrideInfo(1, 1, 0, 0))
.set_name("logits/convolution")
@@ -129,7 +135,8 @@
CommonGraphParams common_params;
Stream graph;
- void add_residual_block(const std::string &data_path, const std::string &name, unsigned int base_depth, unsigned int num_units, unsigned int stride)
+ void add_residual_block(const std::string &data_path, const std::string &name, DataLayout weights_layout,
+ unsigned int base_depth, unsigned int num_units, unsigned int stride)
{
for(unsigned int i = 0; i < num_units; ++i)
{
@@ -151,7 +158,7 @@
SubStream right(graph);
right << ConvolutionLayer(
1U, 1U, base_depth,
- get_weights_accessor(data_path, unit_path + "conv1_weights.npy"),
+ get_weights_accessor(data_path, unit_path + "conv1_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(unit_name + "conv1/convolution")
@@ -166,7 +173,7 @@
<< ConvolutionLayer(
3U, 3U, base_depth,
- get_weights_accessor(data_path, unit_path + "conv2_weights.npy"),
+ get_weights_accessor(data_path, unit_path + "conv2_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(middle_stride, middle_stride, 1, 1))
.set_name(unit_name + "conv2/convolution")
@@ -181,7 +188,7 @@
<< ConvolutionLayer(
1U, 1U, base_depth * 4,
- get_weights_accessor(data_path, unit_path + "conv3_weights.npy"),
+ get_weights_accessor(data_path, unit_path + "conv3_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(unit_name + "conv3/convolution")
@@ -198,7 +205,7 @@
SubStream left(graph);
left << ConvolutionLayer(
1U, 1U, base_depth * 4,
- get_weights_accessor(data_path, unit_path + "shortcut_weights.npy"),
+ get_weights_accessor(data_path, unit_path + "shortcut_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(1, 1, 0, 0))
.set_name(unit_name + "shortcut/convolution")
diff --git a/examples/graph_resnext50.cpp b/examples/graph_resnext50.cpp
index c0e9b9f..e7ef013 100644
--- a/examples/graph_resnext50.cpp
+++ b/examples/graph_resnext50.cpp
@@ -60,7 +60,6 @@
// Checks
ARM_COMPUTE_EXIT_ON_MSG(arm_compute::is_data_type_quantized_asymmetric(common_params.data_type), "Unsupported data type!");
- ARM_COMPUTE_EXIT_ON_MSG(common_params.data_layout == DataLayout::NHWC, "Unsupported data layout!");
// Print parameter values
std::cout << common_params << std::endl;
@@ -68,26 +67,32 @@
// Get trainable parameters data path
std::string data_path = common_params.data_path;
+ // Create input descriptor
+ const TensorShape tensor_shape = permute_shape(TensorShape(224U, 224U, 3U, 1U), DataLayout::NCHW, common_params.data_layout);
+ TensorDescriptor input_descriptor = TensorDescriptor(tensor_shape, common_params.data_type).set_layout(common_params.data_layout);
+
+ // Set weights trained layout
+ const DataLayout weights_layout = DataLayout::NCHW;
+
graph << common_params.target
<< common_params.fast_math_hint
- << InputLayer(TensorDescriptor(TensorShape(224U, 224U, 3U, 1U), common_params.data_type),
- get_input_accessor(common_params))
+ << InputLayer(input_descriptor, get_input_accessor(common_params))
<< ScaleLayer(get_weights_accessor(data_path, "/cnn_data/resnext50_model/bn_data_mul.npy"),
get_weights_accessor(data_path, "/cnn_data/resnext50_model/bn_data_add.npy"))
.set_name("bn_data/Scale")
<< ConvolutionLayer(
7U, 7U, 64U,
- get_weights_accessor(data_path, "/cnn_data/resnext50_model/conv0_weights.npy"),
+ get_weights_accessor(data_path, "/cnn_data/resnext50_model/conv0_weights.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/resnext50_model/conv0_biases.npy"),
PadStrideInfo(2, 2, 2, 3, 2, 3, DimensionRoundingType::FLOOR))
.set_name("conv0/Convolution")
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv0/Relu")
<< PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 1, 0, 1, DimensionRoundingType::FLOOR))).set_name("pool0");
- add_residual_block(data_path, /*ofm*/ 256, /*stage*/ 1, /*num_unit*/ 3, /*stride_conv_unit1*/ 1);
- add_residual_block(data_path, 512, 2, 4, 2);
- add_residual_block(data_path, 1024, 3, 6, 2);
- add_residual_block(data_path, 2048, 4, 3, 2);
+ add_residual_block(data_path, weights_layout, /*ofm*/ 256, /*stage*/ 1, /*num_unit*/ 3, /*stride_conv_unit1*/ 1);
+ add_residual_block(data_path, weights_layout, 512, 2, 4, 2);
+ add_residual_block(data_path, weights_layout, 1024, 3, 6, 2);
+ add_residual_block(data_path, weights_layout, 2048, 4, 3, 2);
graph << PoolingLayer(PoolingLayerInfo(PoolingType::AVG)).set_name("pool1")
<< FlattenLayer().set_name("predictions/Reshape")
@@ -116,7 +121,8 @@
CommonGraphParams common_params;
Stream graph;
- void add_residual_block(const std::string &data_path, unsigned int base_depth, unsigned int stage, unsigned int num_units, unsigned int stride_conv_unit1)
+ void add_residual_block(const std::string &data_path, DataLayout weights_layout,
+ unsigned int base_depth, unsigned int stage, unsigned int num_units, unsigned int stride_conv_unit1)
{
for(unsigned int i = 0; i < num_units; ++i)
{
@@ -137,7 +143,7 @@
SubStream right(graph);
right << ConvolutionLayer(
1U, 1U, base_depth / 2,
- get_weights_accessor(data_path, unit_path + "conv1_weights.npy"),
+ get_weights_accessor(data_path, unit_path + "conv1_weights.npy", weights_layout),
get_weights_accessor(data_path, unit_path + "conv1_biases.npy"),
PadStrideInfo(1, 1, 0, 0))
.set_name(unit_name + "conv1/convolution")
@@ -145,7 +151,7 @@
<< ConvolutionLayer(
3U, 3U, base_depth / 2,
- get_weights_accessor(data_path, unit_path + "conv2_weights.npy"),
+ get_weights_accessor(data_path, unit_path + "conv2_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
pad_grouped_conv, 32)
.set_name(unit_name + "conv2/convolution")
@@ -156,7 +162,7 @@
<< ConvolutionLayer(
1U, 1U, base_depth,
- get_weights_accessor(data_path, unit_path + "conv3_weights.npy"),
+ get_weights_accessor(data_path, unit_path + "conv3_weights.npy", weights_layout),
get_weights_accessor(data_path, unit_path + "conv3_biases.npy"),
PadStrideInfo(1, 1, 0, 0))
.set_name(unit_name + "conv3/convolution");
@@ -166,7 +172,7 @@
{
left << ConvolutionLayer(
1U, 1U, base_depth,
- get_weights_accessor(data_path, unit_path + "sc_weights.npy"),
+ get_weights_accessor(data_path, unit_path + "sc_weights.npy", weights_layout),
std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
PadStrideInfo(stride_conv_unit1, stride_conv_unit1, 0, 0))
.set_name(unit_name + "sc/convolution")
diff --git a/examples/graph_squeezenet.cpp b/examples/graph_squeezenet.cpp
index 9439ab4..b539a9b 100644
--- a/examples/graph_squeezenet.cpp
+++ b/examples/graph_squeezenet.cpp
@@ -60,7 +60,6 @@
// Checks
ARM_COMPUTE_EXIT_ON_MSG(arm_compute::is_data_type_quantized_asymmetric(common_params.data_type), "Unsupported data type!");
- ARM_COMPUTE_EXIT_ON_MSG(common_params.data_layout == DataLayout::NHWC, "Unsupported data layout!");
// Print parameter values
std::cout << common_params << std::endl;
@@ -72,78 +71,84 @@
const std::array<float, 3> mean_rgb{ { 122.68f, 116.67f, 104.01f } };
std::unique_ptr<IPreprocessor> preprocessor = arm_compute::support::cpp14::make_unique<CaffePreproccessor>(mean_rgb);
+ // Create input descriptor
+ const TensorShape tensor_shape = permute_shape(TensorShape(224U, 224U, 3U, 1U), DataLayout::NCHW, common_params.data_layout);
+ TensorDescriptor input_descriptor = TensorDescriptor(tensor_shape, common_params.data_type).set_layout(common_params.data_layout);
+
+ // Set weights trained layout
+ const DataLayout weights_layout = DataLayout::NCHW;
+
graph << common_params.target
<< common_params.fast_math_hint
- << InputLayer(TensorDescriptor(TensorShape(224U, 224U, 3U, 1U), common_params.data_type),
- get_input_accessor(common_params, std::move(preprocessor)))
+ << InputLayer(input_descriptor, get_input_accessor(common_params, std::move(preprocessor)))
<< ConvolutionLayer(
7U, 7U, 96U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv1_b.npy"),
PadStrideInfo(2, 2, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
<< ConvolutionLayer(
1U, 1U, 16U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire2_squeeze1x1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire2_squeeze1x1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire2_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- graph << get_expand_fire_node(data_path, "fire2", 64U, 64U);
+ graph << get_expand_fire_node(data_path, "fire2", weights_layout, 64U, 64U);
graph << ConvolutionLayer(
1U, 1U, 16U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire3_squeeze1x1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire3_squeeze1x1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire3_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- graph << get_expand_fire_node(data_path, "fire3", 64U, 64U);
+ graph << get_expand_fire_node(data_path, "fire3", weights_layout, 64U, 64U);
graph << ConvolutionLayer(
1U, 1U, 32U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire4_squeeze1x1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire4_squeeze1x1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire4_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- graph << get_expand_fire_node(data_path, "fire4", 128U, 128U);
+ graph << get_expand_fire_node(data_path, "fire4", weights_layout, 128U, 128U);
graph << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
<< ConvolutionLayer(
1U, 1U, 32U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire5_squeeze1x1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire5_squeeze1x1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire5_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- graph << get_expand_fire_node(data_path, "fire5", 128U, 128U);
+ graph << get_expand_fire_node(data_path, "fire5", weights_layout, 128U, 128U);
graph << ConvolutionLayer(
1U, 1U, 48U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire6_squeeze1x1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire6_squeeze1x1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire6_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- graph << get_expand_fire_node(data_path, "fire6", 192U, 192U);
+ graph << get_expand_fire_node(data_path, "fire6", weights_layout, 192U, 192U);
graph << ConvolutionLayer(
1U, 1U, 48U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire7_squeeze1x1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire7_squeeze1x1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire7_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- graph << get_expand_fire_node(data_path, "fire7", 192U, 192U);
+ graph << get_expand_fire_node(data_path, "fire7", weights_layout, 192U, 192U);
graph << ConvolutionLayer(
1U, 1U, 64U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire8_squeeze1x1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire8_squeeze1x1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire8_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- graph << get_expand_fire_node(data_path, "fire8", 256U, 256U);
+ graph << get_expand_fire_node(data_path, "fire8", weights_layout, 256U, 256U);
graph << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
<< ConvolutionLayer(
1U, 1U, 64U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire9_squeeze1x1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire9_squeeze1x1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire9_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- graph << get_expand_fire_node(data_path, "fire9", 256U, 256U);
+ graph << get_expand_fire_node(data_path, "fire9", weights_layout, 256U, 256U);
graph << ConvolutionLayer(
1U, 1U, 1000U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv10_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv10_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv10_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
@@ -174,13 +179,14 @@
CommonGraphParams common_params;
Stream graph;
- BranchLayer get_expand_fire_node(const std::string &data_path, std::string &¶m_path, unsigned int expand1_filt, unsigned int expand3_filt)
+ BranchLayer get_expand_fire_node(const std::string &data_path, std::string &¶m_path, DataLayout weights_layout,
+ unsigned int expand1_filt, unsigned int expand3_filt)
{
std::string total_path = "/cnn_data/squeezenet_v1.0_model/" + param_path + "_";
SubStream i_a(graph);
i_a << ConvolutionLayer(
1U, 1U, expand1_filt,
- get_weights_accessor(data_path, total_path + "expand1x1_w.npy"),
+ get_weights_accessor(data_path, total_path + "expand1x1_w.npy", weights_layout),
get_weights_accessor(data_path, total_path + "expand1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
@@ -188,7 +194,7 @@
SubStream i_b(graph);
i_b << ConvolutionLayer(
3U, 3U, expand3_filt,
- get_weights_accessor(data_path, total_path + "expand3x3_w.npy"),
+ get_weights_accessor(data_path, total_path + "expand3x3_w.npy", weights_layout),
get_weights_accessor(data_path, total_path + "expand3x3_b.npy"),
PadStrideInfo(1, 1, 1, 1))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
diff --git a/examples/graph_squeezenet_v1_1.cpp b/examples/graph_squeezenet_v1_1.cpp
index ba7ee77..c0b5ff2 100644
--- a/examples/graph_squeezenet_v1_1.cpp
+++ b/examples/graph_squeezenet_v1_1.cpp
@@ -60,7 +60,6 @@
// Checks
ARM_COMPUTE_EXIT_ON_MSG(arm_compute::is_data_type_quantized_asymmetric(common_params.data_type), "Unsupported data type!");
- ARM_COMPUTE_EXIT_ON_MSG(common_params.data_layout == DataLayout::NHWC, "Unsupported data layout!");
// Print parameter values
std::cout << common_params << std::endl;
@@ -72,80 +71,86 @@
const std::array<float, 3> mean_rgb{ { 122.68f, 116.67f, 104.01f } };
std::unique_ptr<IPreprocessor> preprocessor = arm_compute::support::cpp14::make_unique<CaffePreproccessor>(mean_rgb);
+ // Create input descriptor
+ const TensorShape tensor_shape = permute_shape(TensorShape(224U, 224U, 3U, 1U), DataLayout::NCHW, common_params.data_layout);
+ TensorDescriptor input_descriptor = TensorDescriptor(tensor_shape, common_params.data_type).set_layout(common_params.data_layout);
+
+ // Set weights trained layout
+ const DataLayout weights_layout = DataLayout::NCHW;
+
graph << common_params.target
<< common_params.fast_math_hint
- << InputLayer(TensorDescriptor(TensorShape(227U, 227U, 3U, 1U), common_params.data_type),
- get_input_accessor(common_params, std::move(preprocessor)))
- << ConvolutionMethod::DIRECT
+ << InputLayer(input_descriptor, get_input_accessor(common_params, std::move(preprocessor)))
+ << ConvolutionMethod::Direct
<< ConvolutionLayer(
3U, 3U, 64U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/conv1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/conv1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/conv1_b.npy"),
PadStrideInfo(2, 2, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
<< PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
- << ConvolutionMethod::DEFAULT
+ << ConvolutionMethod::Default
<< ConvolutionLayer(
1U, 1U, 16U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire2_squeeze1x1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire2_squeeze1x1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire2_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- graph << get_expand_fire_node(data_path, "fire2", 64U, 64U);
+ graph << get_expand_fire_node(data_path, "fire2", weights_layout, 64U, 64U);
graph << ConvolutionLayer(
1U, 1U, 16U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire3_squeeze1x1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire3_squeeze1x1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire3_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- graph << get_expand_fire_node(data_path, "fire3", 64U, 64U);
+ graph << get_expand_fire_node(data_path, "fire3", weights_layout, 64U, 64U);
graph << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
<< ConvolutionLayer(
1U, 1U, 32U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire4_squeeze1x1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire4_squeeze1x1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire4_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- graph << get_expand_fire_node(data_path, "fire4", 128U, 128U);
+ graph << get_expand_fire_node(data_path, "fire4", weights_layout, 128U, 128U);
graph << ConvolutionLayer(
1U, 1U, 32U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire5_squeeze1x1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire5_squeeze1x1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire5_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- graph << get_expand_fire_node(data_path, "fire5", 128U, 128U);
+ graph << get_expand_fire_node(data_path, "fire5", weights_layout, 128U, 128U);
graph << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
<< ConvolutionLayer(
1U, 1U, 48U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire6_squeeze1x1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire6_squeeze1x1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire6_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- graph << get_expand_fire_node(data_path, "fire6", 192U, 192U);
+ graph << get_expand_fire_node(data_path, "fire6", weights_layout, 192U, 192U);
graph << ConvolutionLayer(
1U, 1U, 48U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire7_squeeze1x1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire7_squeeze1x1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire7_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- graph << get_expand_fire_node(data_path, "fire7", 192U, 192U);
+ graph << get_expand_fire_node(data_path, "fire7", weights_layout, 192U, 192U);
graph << ConvolutionLayer(
1U, 1U, 64U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire8_squeeze1x1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire8_squeeze1x1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire8_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- graph << get_expand_fire_node(data_path, "fire8", 256U, 256U);
+ graph << get_expand_fire_node(data_path, "fire8", weights_layout, 256U, 256U);
graph << ConvolutionLayer(
1U, 1U, 64U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire9_squeeze1x1_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire9_squeeze1x1_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire9_squeeze1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
- graph << get_expand_fire_node(data_path, "fire9", 256U, 256U);
+ graph << get_expand_fire_node(data_path, "fire9", weights_layout, 256U, 256U);
graph << ConvolutionLayer(
1U, 1U, 1000U,
- get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/conv10_w.npy"),
+ get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/conv10_w.npy", weights_layout),
get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/conv10_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
@@ -176,13 +181,14 @@
CommonGraphParams common_params;
Stream graph;
- BranchLayer get_expand_fire_node(const std::string &data_path, std::string &¶m_path, unsigned int expand1_filt, unsigned int expand3_filt)
+ BranchLayer get_expand_fire_node(const std::string &data_path, std::string &¶m_path, DataLayout weights_layout,
+ unsigned int expand1_filt, unsigned int expand3_filt)
{
std::string total_path = "/cnn_data/squeezenet_v1_1_model/" + param_path + "_";
SubStream i_a(graph);
i_a << ConvolutionLayer(
1U, 1U, expand1_filt,
- get_weights_accessor(data_path, total_path + "expand1x1_w.npy"),
+ get_weights_accessor(data_path, total_path + "expand1x1_w.npy", weights_layout),
get_weights_accessor(data_path, total_path + "expand1x1_b.npy"),
PadStrideInfo(1, 1, 0, 0))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
@@ -190,7 +196,7 @@
SubStream i_b(graph);
i_b << ConvolutionLayer(
3U, 3U, expand3_filt,
- get_weights_accessor(data_path, total_path + "expand3x3_w.npy"),
+ get_weights_accessor(data_path, total_path + "expand3x3_w.npy", weights_layout),
get_weights_accessor(data_path, total_path + "expand3x3_b.npy"),
PadStrideInfo(1, 1, 1, 1))
<< ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));