COMPMID-1460 Create Yolo v3 graph example

Adding darknet53, still need to add YOLO layer.

Change-Id: I0feba46ea850c9107f9e7ea456336e87a0bf9b27
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/147610
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

1 files changed, 267 insertions, 0 deletions

diff --git a/examples/graph_yolov3.cpp b/examples/graph_yolov3.cpp
new file mode 100644
index 0000000000..d2f7ae5e59
--- /dev/null
+++ b/examples/graph_yolov3.cpp
@@ -0,0 +1,267 @@
+/*
+ * Copyright (c) 2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/graph.h"
+#include "support/ToolchainSupport.h"
+#include "utils/CommonGraphOptions.h"
+#include "utils/GraphUtils.h"
+#include "utils/Utils.h"
+
+using namespace arm_compute::utils;
+using namespace arm_compute::graph::frontend;
+using namespace arm_compute::graph_utils;
+
+/** Example demonstrating how to implement YOLOv3 network using the Compute Library's graph API
+ *
+ * @param[in] argc Number of arguments
+ * @param[in] argv Arguments
+ */
+class GraphYOLOv3Example : public Example
+{
+public:
+    GraphYOLOv3Example()
+        : cmd_parser(), common_opts(cmd_parser), common_params(), graph(0, "YOLOv3")
+    {
+    }
+    bool do_setup(int argc, char **argv) override
+    {
+        // Parse arguments
+        cmd_parser.parse(argc, argv);
+
+        // Consume common parameters
+        common_params = consume_common_graph_parameters(common_opts);
+
+        // Return when help menu is requested
+        if(common_params.help)
+        {
+            cmd_parser.print_help(argv[0]);
+            return false;
+        }
+
+        // Checks
+        ARM_COMPUTE_EXIT_ON_MSG(arm_compute::is_data_type_quantized_asymmetric(common_params.data_type), "QASYMM8 not supported for this graph");
+
+        // Print parameter values
+        std::cout << common_params << std::endl;
+
+        // Get trainable parameters data path
+        std::string data_path = common_params.data_path;
+
+        // Create a preprocessor object
+        std::unique_ptr<IPreprocessor> preprocessor = arm_compute::support::cpp14::make_unique<TFPreproccessor>(0.f);
+
+        // Create input descriptor
+        const TensorShape tensor_shape     = permute_shape(TensorShape(608U, 608U, 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(input_descriptor, get_input_accessor(common_params, std::move(preprocessor), false))
+              // Layer 1
+              << ConvolutionLayer(
+                  3U, 3U, 32U,
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_1_w.npy", weights_layout),
+                  std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
+                  PadStrideInfo(1, 1, 1, 1))
+              .set_name("conv2d_1")
+              << BatchNormalizationLayer(
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_1_mean.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_1_var.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_1_gamma.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_1_beta.npy"),
+                  0.000001f)
+              .set_name("conv2d_1/BatchNorm")
+              << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_1/LeakyRelu")
+
+              // Layer 2
+              << ConvolutionLayer(
+                  3U, 3U, 64U,
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_2_w.npy", weights_layout),
+                  std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
+                  PadStrideInfo(2, 2, 1, 1))
+              .set_name("conv2d_2")
+              << BatchNormalizationLayer(
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_2_mean.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_2_var.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_2_gamma.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_2_beta.npy"),
+                  0.000001f)
+              .set_name("conv2d_2/BatchNorm")
+              << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_2/LeakyRelu");
+        darknet53_block(data_path, "3", weights_layout, 32U);
+        graph << ConvolutionLayer(
+                  3U, 3U, 128U,
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_5_w.npy", weights_layout),
+                  std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
+                  PadStrideInfo(2, 2, 1, 1))
+              .set_name("conv2d_5")
+              << BatchNormalizationLayer(
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_5_mean.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_5_var.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_5_gamma.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_5_beta.npy"),
+                  0.000001f)
+              .set_name("conv2d_5/BatchNorm")
+              << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_5/LeakyRelu");
+        darknet53_block(data_path, "6", weights_layout, 64U);
+        darknet53_block(data_path, "8", weights_layout, 64U);
+        graph << ConvolutionLayer(
+                  3U, 3U, 256U,
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_10_w.npy", weights_layout),
+                  std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
+                  PadStrideInfo(2, 2, 1, 1))
+              .set_name("conv2d_10")
+              << BatchNormalizationLayer(
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_10_mean.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_10_var.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_10_gamma.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_10_beta.npy"),
+                  0.000001f)
+              .set_name("conv2d_10/BatchNorm")
+              << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_10/LeakyRelu");
+        darknet53_block(data_path, "11", weights_layout, 128U);
+        darknet53_block(data_path, "13", weights_layout, 128U);
+        darknet53_block(data_path, "15", weights_layout, 128U);
+        darknet53_block(data_path, "17", weights_layout, 128U);
+        darknet53_block(data_path, "19", weights_layout, 128U);
+        darknet53_block(data_path, "21", weights_layout, 128U);
+        darknet53_block(data_path, "23", weights_layout, 128U);
+        darknet53_block(data_path, "25", weights_layout, 128U);
+        graph << ConvolutionLayer(
+                  3U, 3U, 512U,
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_27_w.npy", weights_layout),
+                  std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
+                  PadStrideInfo(2, 2, 1, 1))
+              .set_name("conv2d_27")
+              << BatchNormalizationLayer(
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_27_mean.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_27_var.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_27_gamma.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_27_beta.npy"),
+                  0.000001f)
+              .set_name("conv2d_27/BatchNorm")
+              << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_27/LeakyRelu");
+        darknet53_block(data_path, "28", weights_layout, 256U);
+        darknet53_block(data_path, "30", weights_layout, 256U);
+        darknet53_block(data_path, "32", weights_layout, 256U);
+        darknet53_block(data_path, "34", weights_layout, 256U);
+        darknet53_block(data_path, "36", weights_layout, 256U);
+        darknet53_block(data_path, "38", weights_layout, 256U);
+        darknet53_block(data_path, "40", weights_layout, 256U);
+        darknet53_block(data_path, "42", weights_layout, 256U);
+        graph << ConvolutionLayer(
+                  3U, 3U, 1024U,
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_44_w.npy", weights_layout),
+                  std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
+                  PadStrideInfo(2, 2, 1, 1))
+              .set_name("conv2d_44")
+              << BatchNormalizationLayer(
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_44_mean.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_44_var.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_44_gamma.npy"),
+                  get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_44_beta.npy"),
+                  0.000001f)
+              .set_name("conv2d_44/BatchNorm")
+              << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_44/LeakyRelu");
+        darknet53_block(data_path, "45", weights_layout, 512U);
+        darknet53_block(data_path, "47", weights_layout, 512U);
+        darknet53_block(data_path, "49", weights_layout, 512U);
+        darknet53_block(data_path, "51", weights_layout, 512U);
+        graph << OutputLayer(get_output_accessor(common_params, 5));
+
+        // Finalize graph
+        GraphConfig config;
+        config.num_threads = common_params.threads;
+        config.use_tuner   = common_params.enable_tuner;
+        config.tuner_file  = common_params.tuner_file;
+
+        graph.finalize(common_params.target, config);
+
+        return true;
+    }
+    void do_run() override
+    {
+        // Run graph
+        graph.run();
+    }
+
+private:
+    CommandLineParser  cmd_parser;
+    CommonGraphOptions common_opts;
+    CommonGraphParams  common_params;
+    Stream             graph;
+
+    void darknet53_block(const std::string &data_path, std::string &&param_path, DataLayout weights_layout,
+                         unsigned int filter_size)
+    {
+        std::string total_path  = "/cnn_data/yolov3_model/";
+        std::string param_path2 = std::to_string(std::stoi(param_path) + 1);
+        SubStream   i_a(graph);
+        SubStream   i_b(graph);
+        i_a << ConvolutionLayer(
+                1U, 1U, filter_size,
+                get_weights_accessor(data_path, total_path + "conv2d_" + param_path + "_w.npy", weights_layout),
+                std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
+                PadStrideInfo(1, 1, 0, 0))
+            << BatchNormalizationLayer(
+                get_weights_accessor(data_path, total_path + "batch_normalization_" + param_path + "_mean.npy"),
+                get_weights_accessor(data_path, total_path + "batch_normalization_" + param_path + "_var.npy"),
+                get_weights_accessor(data_path, total_path + "batch_normalization_" + param_path + "_gamma.npy"),
+                get_weights_accessor(data_path, total_path + "batch_normalization_" + param_path + "_beta.npy"),
+                0.000001f)
+            .set_name("conv2d" + param_path + "/BatchNorm")
+            << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d" + param_path + "/LeakyRelu")
+            << ConvolutionLayer(
+                3U, 3U, filter_size * 2,
+                get_weights_accessor(data_path, total_path + "conv2d_" + param_path2 + "_w.npy", weights_layout),
+                std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
+                PadStrideInfo(1, 1, 1, 1))
+            << BatchNormalizationLayer(
+                get_weights_accessor(data_path, total_path + "batch_normalization_" + param_path2 + "_mean.npy"),
+                get_weights_accessor(data_path, total_path + "batch_normalization_" + param_path2 + "_var.npy"),
+                get_weights_accessor(data_path, total_path + "batch_normalization_" + param_path2 + "_gamma.npy"),
+                get_weights_accessor(data_path, total_path + "batch_normalization_" + param_path2 + "_beta.npy"),
+                0.000001f)
+            .set_name("conv2d" + param_path2 + "/BatchNorm")
+            << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d" + param_path2 + "/LeakyRelu");
+
+        graph << EltwiseLayer(std::move(i_a), std::move(i_b), EltwiseOperation::Add);
+    }
+};
+
+/** Main program for YOLOv3
+ *
+ * @note To list all the possible arguments execute the binary appended with the --help option
+ *
+ * @param[in] argc Number of arguments
+ * @param[in] argv Arguments
+ *
+ * @return Return code
+ */
+int main(int argc, char **argv)
+{
+    return arm_compute::utils::run_example<GraphYOLOv3Example>(argc, argv);
+}