/* * Copyright (c) 2017-2018 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 "benchmark-common.h" #if defined(ARMCL_18_05_PLUS) using namespace arm_compute::graph::frontend; #endif // What's new in SqueezeNet v1.1? // Conv 1: 64 filters (3x3) // Pool 1 3 5 insted of 1 4 8 void run_squeezenet11() { std::string data_path; /* Path to the trainable data */ std::string image; /* Image data */ std::string label; /* Label data */ auto target_hint = get_target_hint(); auto convolution_method = get_convolution_method(); GRAPH(graph, "SqueezeNetV1.1"); #if defined(ARMCL_18_05_PLUS) #define SUB_GRAPH(var_name) SubStream var_name(graph) #else #define SUB_GRAPH(var_name) SubGraph var_name #endif auto get_expand_fire_node = [&](const std::string &data_path, std::string &¶m_path, unsigned int expand1_filt, unsigned int expand3_filt) -> BranchLayer { std::string total_path = "/cnn_data/squeezenet_v1.1_model/" + param_path + "_"; SUB_GRAPH(i_a); 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_b.npy"), PadStrideInfo(1, 1, 0, 0)) << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)); SUB_GRAPH(i_b); 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_b.npy"), PadStrideInfo(1, 1, 1, 1)) << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)); return BranchLayer(BranchMergeMethod::DEPTH_CONCATENATE, std::move(i_a), std::move(i_b)); }; xopenme_clock_start(X_TIMER_SETUP); graph << target_hint << convolution_method << make_input_layer(image, 224U, 224U, 3U, { 122.68f, 116.67f, 104.01f }) << 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_b.npy"), PadStrideInfo(2, 2, 0, 0)) // Conv 1 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) // RELU 1 << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))) // Pool 1 (Fire1) << 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_b.npy"), PadStrideInfo(1, 1, 0, 0)) // Conv 2 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) // RELU 2 << get_expand_fire_node(data_path, "fire2", 64U, 64U) // Expand and Concat 1 << 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_b.npy"), PadStrideInfo(1, 1, 0, 0)) // Conv 3 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) // Relu 3 << get_expand_fire_node(data_path, "fire3", 64U, 64U) // Expand and Concat 2 << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))) // Pool 2 (fire 3) << 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_b.npy"), PadStrideInfo(1, 1, 0, 0)) // Conv 4 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) // Relu 4 << get_expand_fire_node(data_path, "fire4", 128U, 128U) // Expand and Concat 3 << 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_b.npy"), PadStrideInfo(1, 1, 0, 0)) // Conv 4 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) // Relu 4 << get_expand_fire_node(data_path, "fire5", 128U, 128U) // Expand Concat 4 fire 5 << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))) // Pool 3 (Fire 5) << 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_b.npy"), PadStrideInfo(1, 1, 0, 0)) << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) << get_expand_fire_node(data_path, "fire6", 192U, 192U) << 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_b.npy"), PadStrideInfo(1, 1, 0, 0)) << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) << get_expand_fire_node(data_path, "fire7", 192U, 192U) << 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_b.npy"), PadStrideInfo(1, 1, 0, 0)) << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) << get_expand_fire_node(data_path, "fire8", 256U, 256U) << 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_b.npy"), PadStrideInfo(1, 1, 0, 0)) << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) << get_expand_fire_node(data_path, "fire9", 256U, 256U) << 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_b.npy"), PadStrideInfo(1, 1, 0, 0)) << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) << PoolingLayer(PoolingLayerInfo(PoolingType::AVG)) << FlattenLayer() << SoftmaxLayer() << make_output_layer(label); #if defined(ARMCL_18_05_PLUS) // Finalize graph GraphConfig config{}; graph.finalize(target_hint, config); #endif xopenme_clock_end(X_TIMER_SETUP); xopenme_clock_start(X_TIMER_TEST); graph.run(); xopenme_clock_end(X_TIMER_TEST); }