| /* |
| * Copyright (c) 2018-2021 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 */ |
| 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); |
| cmd_parser.validate(); |
| |
| // 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 = std::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)); |
| std::pair<SubStream, SubStream> intermediate_layers = darknet53(data_path, weights_layout); |
| graph << ConvolutionLayer( |
| 1U, 1U, 512U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_53_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 0, 0)) |
| .set_name("conv2d_53") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_53_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_53_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_53_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_53_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_53/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_53/LeakyRelu") |
| << ConvolutionLayer( |
| 3U, 3U, 1024U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_54_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 1, 1)) |
| .set_name("conv2d_54") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_54_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_54_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_54_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_54_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_54/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_54/LeakyRelu") |
| << ConvolutionLayer( |
| 1U, 1U, 512U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_55_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 0, 0)) |
| .set_name("conv2d_55") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_55_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_55_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_55_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_55_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_55/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_55/LeakyRelu") |
| << ConvolutionLayer( |
| 3U, 3U, 1024U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_56_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 1, 1)) |
| .set_name("conv2d_56") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_56_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_56_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_56_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_56_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_56/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_56/LeakyRelu") |
| << ConvolutionLayer( |
| 1U, 1U, 512U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_57_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 0, 0)) |
| .set_name("conv2d_57") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_57_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_57_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_57_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_57_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_57/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_57/LeakyRelu"); |
| SubStream route_1(graph); |
| graph << ConvolutionLayer( |
| 3U, 3U, 1024U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_58_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 1, 1)) |
| .set_name("conv2d_58") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_58_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_58_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_58_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_58_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_58/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_58/LeakyRelu") |
| << ConvolutionLayer( |
| 1U, 1U, 255U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_59_w.npy", weights_layout), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_59_b.npy", weights_layout), |
| PadStrideInfo(1, 1, 0, 0)) |
| .set_name("conv2d_59") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LINEAR, 1.f)).set_name("conv2d_59/Linear") |
| << YOLOLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC, 0.1f)).set_name("Yolo1") |
| << OutputLayer(get_output_accessor(common_params, 5)); |
| route_1 << ConvolutionLayer( |
| 1U, 1U, 256U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_60_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 0, 0)) |
| .set_name("conv2d_60") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_59_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_59_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_59_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_59_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_59/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_60/LeakyRelu") |
| << ResizeLayer(InterpolationPolicy::NEAREST_NEIGHBOR, 2, 2).set_name("Upsample_60"); |
| SubStream concat_1(route_1); |
| concat_1 << ConcatLayer(std::move(route_1), std::move(intermediate_layers.second)).set_name("Route1") |
| << ConvolutionLayer( |
| 1U, 1U, 256U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_61_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 0, 0)) |
| .set_name("conv2d_61") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_60_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_60_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_60_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_60_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_60/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_61/LeakyRelu") |
| << ConvolutionLayer( |
| 3U, 3U, 512U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_62_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 1, 1)) |
| .set_name("conv2d_62") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_61_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_61_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_61_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_61_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_61/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_62/LeakyRelu") |
| << ConvolutionLayer( |
| 1U, 1U, 256U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_63_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 0, 0)) |
| .set_name("conv2d_63") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_62_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_62_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_62_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_62_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_62/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_63/LeakyRelu") |
| << ConvolutionLayer( |
| 3U, 3U, 512U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_64_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 1, 1)) |
| .set_name("conv2d_64") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_63_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_63_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_63_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_63_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_63/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_64/LeakyRelu") |
| << ConvolutionLayer( |
| 1U, 1U, 256U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_65_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 0, 0)) |
| .set_name("conv2d_65") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_64_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_64_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_64_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_64_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_65/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_65/LeakyRelu"); |
| SubStream route_2(concat_1); |
| concat_1 << ConvolutionLayer( |
| 3U, 3U, 512U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_66_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 1, 1)) |
| .set_name("conv2d_66") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_65_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_65_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_65_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_65_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_65/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_66/LeakyRelu") |
| << ConvolutionLayer( |
| 1U, 1U, 255U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_67_w.npy", weights_layout), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_67_b.npy", weights_layout), |
| PadStrideInfo(1, 1, 0, 0)) |
| .set_name("conv2d_67") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LINEAR, 1.f)).set_name("conv2d_67/Linear") |
| << YOLOLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC, 0.1f)).set_name("Yolo2") |
| << OutputLayer(get_output_accessor(common_params, 5)); |
| route_2 << ConvolutionLayer( |
| 1U, 1U, 128U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_68_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 0, 0)) |
| .set_name("conv2d_68") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_66_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_66_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_66_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_66_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_66/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_68/LeakyRelu") |
| << ResizeLayer(InterpolationPolicy::NEAREST_NEIGHBOR, 2, 2).set_name("Upsample_68"); |
| SubStream concat_2(route_2); |
| concat_2 << ConcatLayer(std::move(route_2), std::move(intermediate_layers.first)).set_name("Route2") |
| << ConvolutionLayer( |
| 1U, 1U, 128U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_69_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 0, 0)) |
| .set_name("conv2d_69") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_67_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_67_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_67_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_67_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_67/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_69/LeakyRelu") |
| << ConvolutionLayer( |
| 3U, 3U, 256U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_70_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 1, 1)) |
| .set_name("conv2d_70") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_68_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_68_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_68_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_68_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_68/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_70/LeakyRelu") |
| << ConvolutionLayer( |
| 1U, 1U, 128U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_71_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 0, 0)) |
| .set_name("conv2d_71") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_69_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_69_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_69_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_69_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_69/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_71/LeakyRelu") |
| << ConvolutionLayer( |
| 3U, 3U, 256U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_72_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 1, 1)) |
| .set_name("conv2d_72") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_70_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_70_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_70_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_70_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_70/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_72/LeakyRelu") |
| << ConvolutionLayer( |
| 1U, 1U, 128U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_73_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 0, 0)) |
| .set_name("conv2d_73") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_71_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_71_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_71_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_71_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_71/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_73/LeakyRelu") |
| << ConvolutionLayer( |
| 3U, 3U, 256U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_74_w.npy", weights_layout), |
| std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| PadStrideInfo(1, 1, 1, 1)) |
| .set_name("conv2d_74") |
| << BatchNormalizationLayer( |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_72_mean.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_72_var.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_72_gamma.npy"), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/batch_normalization_72_beta.npy"), |
| 0.000001f) |
| .set_name("conv2d_72/BatchNorm") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LEAKY_RELU, 0.1f)).set_name("conv2d_74/LeakyRelu") |
| << ConvolutionLayer( |
| 1U, 1U, 255U, |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_75_w.npy", weights_layout), |
| get_weights_accessor(data_path, "/cnn_data/yolov3_model/conv2d_75_b.npy", weights_layout), |
| PadStrideInfo(1, 1, 0, 0)) |
| .set_name("conv2d_75") |
| << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LINEAR, 1.f)).set_name("conv2d_75/Linear") |
| << YOLOLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC, 0.1f)).set_name("Yolo3") |
| << 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_mode = common_params.tuner_mode; |
| config.tuner_file = common_params.tuner_file; |
| config.mlgo_file = common_params.mlgo_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; |
| |
| std::pair<SubStream, SubStream> darknet53(const std::string &data_path, DataLayout weights_layout) |
| { |
| graph << 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/Conv2D") |
| << 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") |
| << 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/Conv2D") |
| << 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/Conv2D") |
| << 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/Conv2D") |
| << 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); |
| SubStream layer_36(graph); |
| 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/Conv2D") |
| << 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); |
| SubStream layer_61(graph); |
| 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/Conv2D") |
| << 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); |
| |
| return std::pair<SubStream, SubStream>(layer_36, layer_61); |
| } |
| |
| void darknet53_block(const std::string &data_path, std::string &¶m_path, DataLayout weights_layout, |
| unsigned int filter_size) |
| { |
| std::string total_path = "/cnn_data/yolov3_model/"; |
| std::string param_path2 = arm_compute::support::cpp11::to_string(arm_compute::support::cpp11::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)) |
| .set_name("conv2d_" + param_path + "/Conv2D") |
| << 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)) |
| .set_name("conv2d_" + param_path2 + "/Conv2D") |
| << 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).set_name("").set_name("add_" + param_path + "_" + param_path2); |
| } |
| }; |
| |
| /** Main program for YOLOv3 |
| * |
| * Model is based on: |
| * https://arxiv.org/abs/1804.02767 |
| * "YOLOv3: An Incremental Improvement" |
| * Joseph Redmon, Ali Farhadi |
| * |
| * @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); |
| } |