├── CMakeLists.txt
├── MobileYOLOv3
├── README.md
├── dog-cycle-car.png
├── mobilenet.cpp
├── mobilenetv2_yolov3.bin
└── mobilenetv2_yolov3.param


/CMakeLists.txt:
--------------------------------------------------------------------------------
 1 | cmake_minimum_required(VERSION 3.15)
 2 | project(MobileYOLOv3)
 3 | find_package(OpenCV REQUIRED)
 4 | # find_package(ncnn REQUIRED)
 5 | 
 6 | # set link library and head files of ncnn
 7 | set(NCNN_LIBS "/home/northernjay/ncnn/build/install/lib/libncnn.a")
 8 | set(NCNN_INCLUDE_DIRS "/home/northernjay/ncnn/build/install/include/ncnn")
 9 | 
10 | # configure OpenMP
11 | find_package(OpenMP REQUIRED)
12 | if(OPENMP_FOUND)
13 |     message("OPENMP FOUND")
14 |     set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OpenMP_C_FLAGS}")
15 |     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OpenMP_CXX_FLAGS}")
16 |     set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} ${OpenMP_EXE_LINKER_FLAGS}")
17 | endif()
18 | 
19 | 
20 | include_directories(${NCNN_INCLUDE_DIRS})
21 | #include_directories(${OpenCV_INCLUDE_DIR})
22 | #include_directories(${NCNN_INCLUDE_DIRSS})
23 | 
24 | get_property(dirs DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} PROPERTY INCLUDE_DIRECTORIES)
25 | foreach(dir ${dirs})
26 |   message(STATUS "dir='${dir}'")
27 | endforeach()
28 | 
29 | add_executable(MobileYOLOv3 mobilenet.cpp)
30 | target_link_libraries(MobileYOLOv3 ${OpenCV_LIBS} ${NCNN_LIBS})
31 | 


--------------------------------------------------------------------------------
/MobileYOLOv3:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/northernjay885/NCNN-Mobilenetv2-YOLOv3/06190f606817e7f82b9f5d6027a8f23afc591b00/MobileYOLOv3


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | # NCNN-Mobilenetv2-YOLOv3
 2 | 
 3 | Build a YOLOv3 Convolutional Neural Network for object detection using NCNN framework.
 4 | 
 5 | ## Getting Started
 6 | 
 7 | ```
 8 | git clone https://github.com/northernjay885/NCNN-Mobilenetv2-YOLOv3.git
 9 | ```
10 | 
11 | ### Prerequisites
12 | 
13 | 
14 | Install [CMake](https://cmake.org/download/).
15 | 
16 | ### Installing
17 | 
18 | Build OpenCV using CMake
19 | ```
20 | cmake -D CMAKE_BUILD_TYPE=Release -D CMAKE_INSTALL_PREFIX=/usr/local ..
21 | ```
22 | ```
23 | make -j8 # runs 8 jobs in parallel, depend on your CPU threads
24 | ```
25 | Build NCNN framework following the [guide](https://github.com/Tencent/ncnn).
26 | 
27 | ## Build objects
28 | 
29 | in /build directory
30 | ```
31 | cmake -D CMAKE_BUILD_TYPE=Release ..
32 | ```
33 | 
34 | ## Built With
35 | 
36 | * [Cmake](https://cmake.org/)
37 | * [OpenCV](https://opencv.org/)
38 | * [NCNN](https://github.com/Tencent/ncnn)
39 | 


--------------------------------------------------------------------------------
/dog-cycle-car.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/northernjay885/NCNN-Mobilenetv2-YOLOv3/06190f606817e7f82b9f5d6027a8f23afc591b00/dog-cycle-car.png


--------------------------------------------------------------------------------
/mobilenet.cpp:
--------------------------------------------------------------------------------
  1 | #include <iostream>
  2 | #include <vector>
  3 | #include <string>
  4 | #include <opencv2/core/core.hpp>
  5 | #include <opencv2/highgui/highgui.hpp>
  6 | #include <opencv2/imgproc/imgproc.hpp>
  7 | #include <opencv2/videoio.hpp>
  8 | 
  9 | #include "net.h"
 10 | //#include "mat.h"
 11 | #include "platform.h"
 12 | 
 13 | struct Object
 14 | {
 15 |     cv::Rect_<float> rect;
 16 |     int label;
 17 |     float prob;
 18 | };
 19 | 
 20 | static int detect_yolov3(const cv::Mat& image, std::vector<Object>& objects)
 21 | {
 22 |     const float probThreshold {0.5};
 23 |     //load model and pretrained weights
 24 |     ncnn::Net net;
 25 |     net.load_param("mobilenetv2_yolov3.param");
 26 |     net.load_model("mobilenetv2_yolov3.bin");
 27 | 
 28 |     const int input_size = 352;
 29 | 
 30 |     int img_width = image.cols;
 31 |     int img_height = image.rows;
 32 | 
 33 |     ncnn::Mat ncnnImage = ncnn::Mat::from_pixels_resize(image.data, ncnn::Mat::PIXEL_BGR, img_width, img_height, input_size, input_size);
 34 | 
 35 |     //subtract range mean value, norm to (-1, 1)
 36 |     const float mean_vals[3] = {127.5f, 127.5f, 127.5f};
 37 |     const float norm_vals[3] = {0.007843f, 0.007843f, 0.007843f};
 38 |     ncnnImage.substract_mean_normalize(mean_vals, norm_vals);
 39 | 
 40 |     ncnn::Extractor ex = net.create_extractor();
 41 |     ex.set_num_threads(4);
 42 | 
 43 |     ex.input("data", ncnnImage);
 44 | 
 45 |     ncnn::Mat detectRes;
 46 |     ex.extract("detection_out", detectRes);
 47 | 
 48 |     objects.clear();
 49 | 
 50 |     for(int i = 0; i < detectRes.h; i++)
 51 |     {
 52 |         const float* values = detectRes.row(i);
 53 |         
 54 |         if(values[1] >= probThreshold)
 55 |         {
 56 |             Object object;
 57 |             object.label = values[0];
 58 |             object.prob = values[1];
 59 |             object.rect.x = values[2] * img_width;
 60 |             object.rect.y = values[3] * img_height;
 61 |             object.rect.width = values[4] * img_width - object.rect.x;
 62 |             object.rect.height = values[5] * img_height - object.rect.y;
 63 | 
 64 |             objects.push_back(object);
 65 |         }
 66 | 
 67 |     }
 68 | 
 69 |     return 0;
 70 | 
 71 | }
 72 | 
 73 | static void draw_objects(const cv::Mat& inputImage, const std::vector<Object>& objects)
 74 | {
 75 |     static const char* class_name[] = {"background",
 76 |         "aeroplane", "bicycle", "bird", "boat",
 77 |         "bottle", "bus", "car", "cat", "chair",
 78 |         "cow", "diningtable", "dog", "horse",
 79 |         "motorbike", "person", "pottedplant",
 80 |         "sheep", "sofa", "train", "tvmonitor"};
 81 |     
 82 |     cv::Mat image = inputImage.clone();
 83 | 
 84 |     for(size_t i = 0; i < objects.size(); i++)
 85 |     {
 86 |         const Object& obj = objects[i];
 87 |         cv::rectangle(image, obj.rect, cv::Scalar(50,205,50));
 88 | 
 89 |         std::string text {""};
 90 |         text += class_name[obj.label];
 91 |         text += std::to_string(obj.prob * 100);
 92 |         text += "%";
 93 | 
 94 |         int baseline = 0;
 95 |         cv::Size textSize = cv::getTextSize(text, cv::FONT_HERSHEY_DUPLEX, 0.5, 1, &baseline);
 96 | 
 97 |         int x = obj.rect.x;
 98 |         int y = obj.rect.y;
 99 |         if(y < 0)
100 |             y = 0;
101 |         if(x + textSize.width > image.cols)
102 |             x = image.cols - textSize.width;
103 |         
104 |         //y + textSize.height for bottom left corner
105 |         cv::rectangle(image, cv::Rect(x, y, textSize.width, textSize.height + 5), cv::Scalar(50,205,50), -1);
106 |         cv::putText(image, text, cv::Point(x, y + textSize.height), cv::FONT_HERSHEY_DUPLEX, 0.5, cv::Scalar(255,255,255));
107 |     }
108 | 
109 |     //cv::imshow("image", image);
110 |     //cv::waitKey(0);
111 | }
112 | 
113 | 
114 | int main(int argc, char** argv)
115 | {
116 |     cv::Mat frame;
117 | 
118 |     cv::VideoCapture cap;
119 | 
120 |     // 0 = open default camera
121 |     int deviceID = 0;
122 |     // 0 = autodetect default API
123 |     int apiID = cv::CAP_ANY;
124 | 
125 |     //open camera
126 |     cap.open(deviceID + apiID);
127 | 
128 |     if(!cap.isOpened())
129 |     {
130 |         std::cerr<<"ERROR! Unable to open camera\n";
131 |         return -1;
132 |     }
133 | 
134 |     //--- grab and write loop
135 |     while(true)
136 |     {
137 |         cap.read(frame);
138 |         if(frame.empty())
139 |         {
140 |             std::cerr<<"ERROR! blank frame grabbed\n";
141 |             return -1;
142 |         }
143 | 
144 |         std::vector<Object> objects;
145 | 
146 |         detect_yolov3(frame, objects);
147 |         draw_objects(frame, objects);
148 |         
149 |         cv::imshow("image", frame);
150 |         if(cv::waitKey(5) >= 0)
151 |             break;
152 |     }
153 | 
154 |     //cv::Mat image = cv::imread(imagepath, cv::IMREAD_COLOR);
155 |     // if(image.empty())
156 |     // {
157 |     //     std::cerr<<"no image has been read!"<<std::endl;
158 |     //     return -1;
159 |     // }
160 | 
161 |     //detection
162 |     
163 |     
164 | 
165 |     //draw box
166 |     
167 | 
168 |     return 0;
169 | 
170 | }


--------------------------------------------------------------------------------
/mobilenetv2_yolov3.bin:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/northernjay885/NCNN-Mobilenetv2-YOLOv3/06190f606817e7f82b9f5d6027a8f23afc591b00/mobilenetv2_yolov3.bin


--------------------------------------------------------------------------------
/mobilenetv2_yolov3.param:
--------------------------------------------------------------------------------
  1 | 7767517
  2 | 128 140
  3 | Input            data             0 1 data 0=352 1=352 2=3
  4 | Convolution      conv1            1 1 data conv1 0=32 1=3 2=1 3=2 4=1 5=1 6=864
  5 | ReLU             relu1            1 1 conv1 conv1_relu1
  6 | ConvolutionDepthWise conv2            1 1 conv1_relu1 conv2 0=32 1=3 2=1 3=1 4=1 5=1 6=288 7=32
  7 | ReLU             relu2            1 1 conv2 conv2_relu2
  8 | Convolution      conv3            1 1 conv2_relu2 conv3 0=16 1=1 2=1 3=1 4=0 5=1 6=512
  9 | Convolution      conv4            1 1 conv3 conv4 0=96 1=1 2=1 3=1 4=0 5=1 6=1536
 10 | ReLU             relu3            1 1 conv4 conv4_relu3
 11 | ConvolutionDepthWise conv5            1 1 conv4_relu3 conv5 0=96 1=3 2=1 3=2 4=1 5=1 6=864 7=96
 12 | ReLU             relu4            1 1 conv5 conv5_relu4
 13 | Convolution      conv6            1 1 conv5_relu4 conv6 0=24 1=1 2=1 3=1 4=0 5=1 6=2304
 14 | Split            splitncnn_0      1 2 conv6 conv6_splitncnn_0 conv6_splitncnn_1
 15 | Convolution      conv7            1 1 conv6_splitncnn_1 conv7 0=144 1=1 2=1 3=1 4=0 5=1 6=3456
 16 | ReLU             relu5            1 1 conv7 conv7_relu5
 17 | ConvolutionDepthWise conv8            1 1 conv7_relu5 conv8 0=144 1=3 2=1 3=1 4=1 5=1 6=1296 7=144
 18 | ReLU             relu6            1 1 conv8 conv8_relu6
 19 | Convolution      conv9            1 1 conv8_relu6 conv9 0=24 1=1 2=1 3=1 4=0 5=1 6=3456
 20 | Eltwise          add1             2 1 conv6_splitncnn_0 conv9 add1 0=1 -23301=0
 21 | Convolution      conv10           1 1 add1 conv10 0=144 1=1 2=1 3=1 4=0 5=1 6=3456
 22 | ReLU             relu7            1 1 conv10 conv10_relu7
 23 | ConvolutionDepthWise conv11           1 1 conv10_relu7 conv11 0=144 1=3 2=1 3=2 4=1 5=1 6=1296 7=144
 24 | ReLU             relu8            1 1 conv11 conv11_relu8
 25 | Convolution      conv12           1 1 conv11_relu8 conv12 0=32 1=1 2=1 3=1 4=0 5=1 6=4608
 26 | Split            splitncnn_1      1 2 conv12 conv12_splitncnn_0 conv12_splitncnn_1
 27 | Convolution      conv13           1 1 conv12_splitncnn_1 conv13 0=192 1=1 2=1 3=1 4=0 5=1 6=6144
 28 | ReLU             relu9            1 1 conv13 conv13_relu9
 29 | ConvolutionDepthWise conv14           1 1 conv13_relu9 conv14 0=192 1=3 2=1 3=1 4=1 5=1 6=1728 7=192
 30 | ReLU             relu10           1 1 conv14 conv14_relu10
 31 | Convolution      conv15           1 1 conv14_relu10 conv15 0=32 1=1 2=1 3=1 4=0 5=1 6=6144
 32 | Eltwise          add2             2 1 conv12_splitncnn_0 conv15 add2 0=1 -23301=0
 33 | Split            splitncnn_2      1 2 add2 add2_splitncnn_0 add2_splitncnn_1
 34 | Convolution      conv16           1 1 add2_splitncnn_1 conv16 0=192 1=1 2=1 3=1 4=0 5=1 6=6144
 35 | ReLU             relu11           1 1 conv16 conv16_relu11
 36 | ConvolutionDepthWise conv17           1 1 conv16_relu11 conv17 0=192 1=3 2=1 3=1 4=1 5=1 6=1728 7=192
 37 | ReLU             relu12           1 1 conv17 conv17_relu12
 38 | Convolution      conv18           1 1 conv17_relu12 conv18 0=32 1=1 2=1 3=1 4=0 5=1 6=6144
 39 | Eltwise          add3             2 1 add2_splitncnn_0 conv18 add3 0=1 -23301=0
 40 | Convolution      conv19           1 1 add3 conv19 0=192 1=1 2=1 3=1 4=0 5=1 6=6144
 41 | ReLU             relu13           1 1 conv19 conv19_relu13
 42 | ConvolutionDepthWise conv20           1 1 conv19_relu13 conv20 0=192 1=3 2=1 3=2 4=1 5=1 6=1728 7=192
 43 | ReLU             relu14           1 1 conv20 conv20_relu14
 44 | Convolution      conv21           1 1 conv20_relu14 conv21 0=64 1=1 2=1 3=1 4=0 5=1 6=12288
 45 | Split            splitncnn_3      1 2 conv21 conv21_splitncnn_0 conv21_splitncnn_1
 46 | Convolution      conv22           1 1 conv21_splitncnn_1 conv22 0=384 1=1 2=1 3=1 4=0 5=1 6=24576
 47 | ReLU             relu15           1 1 conv22 conv22_relu15
 48 | ConvolutionDepthWise conv23           1 1 conv22_relu15 conv23 0=384 1=3 2=1 3=1 4=1 5=1 6=3456 7=384
 49 | ReLU             relu16           1 1 conv23 conv23_relu16
 50 | Convolution      conv24           1 1 conv23_relu16 conv24 0=64 1=1 2=1 3=1 4=0 5=1 6=24576
 51 | Eltwise          add4             2 1 conv21_splitncnn_0 conv24 add4 0=1 -23301=0
 52 | Split            splitncnn_4      1 2 add4 add4_splitncnn_0 add4_splitncnn_1
 53 | Convolution      conv25           1 1 add4_splitncnn_1 conv25 0=384 1=1 2=1 3=1 4=0 5=1 6=24576
 54 | ReLU             relu17           1 1 conv25 conv25_relu17
 55 | ConvolutionDepthWise conv26           1 1 conv25_relu17 conv26 0=384 1=3 2=1 3=1 4=1 5=1 6=3456 7=384
 56 | ReLU             relu18           1 1 conv26 conv26_relu18
 57 | Convolution      conv27           1 1 conv26_relu18 conv27 0=64 1=1 2=1 3=1 4=0 5=1 6=24576
 58 | Eltwise          add5             2 1 add4_splitncnn_0 conv27 add5 0=1 -23301=0
 59 | Split            splitncnn_5      1 2 add5 add5_splitncnn_0 add5_splitncnn_1
 60 | Convolution      conv28           1 1 add5_splitncnn_1 conv28 0=384 1=1 2=1 3=1 4=0 5=1 6=24576
 61 | ReLU             relu19           1 1 conv28 conv28_relu19
 62 | ConvolutionDepthWise conv29           1 1 conv28_relu19 conv29 0=384 1=3 2=1 3=1 4=1 5=1 6=3456 7=384
 63 | ReLU             relu20           1 1 conv29 conv29_relu20
 64 | Convolution      conv30           1 1 conv29_relu20 conv30 0=64 1=1 2=1 3=1 4=0 5=1 6=24576
 65 | Eltwise          add6             2 1 add5_splitncnn_0 conv30 add6 0=1 -23301=0
 66 | Convolution      conv31           1 1 add6 conv31 0=384 1=1 2=1 3=1 4=0 5=1 6=24576
 67 | ReLU             relu21           1 1 conv31 conv31_relu21
 68 | ConvolutionDepthWise conv32           1 1 conv31_relu21 conv32 0=384 1=3 2=1 3=1 4=1 5=1 6=3456 7=384
 69 | ReLU             relu22           1 1 conv32 conv32_relu22
 70 | Convolution      conv33           1 1 conv32_relu22 conv33 0=96 1=1 2=1 3=1 4=0 5=1 6=36864
 71 | Split            splitncnn_6      1 2 conv33 conv33_splitncnn_0 conv33_splitncnn_1
 72 | Convolution      conv34           1 1 conv33_splitncnn_1 conv34 0=576 1=1 2=1 3=1 4=0 5=1 6=55296
 73 | ReLU             relu23           1 1 conv34 conv34_relu23
 74 | ConvolutionDepthWise conv35           1 1 conv34_relu23 conv35 0=576 1=3 2=1 3=1 4=1 5=1 6=5184 7=576
 75 | ReLU             relu24           1 1 conv35 conv35_relu24
 76 | Convolution      conv36           1 1 conv35_relu24 conv36 0=96 1=1 2=1 3=1 4=0 5=1 6=55296
 77 | Eltwise          add7             2 1 conv33_splitncnn_0 conv36 add7 0=1 -23301=0
 78 | Split            splitncnn_7      1 2 add7 add7_splitncnn_0 add7_splitncnn_1
 79 | Convolution      conv37           1 1 add7_splitncnn_1 conv37 0=576 1=1 2=1 3=1 4=0 5=1 6=55296
 80 | ReLU             relu25           1 1 conv37 conv37_relu25
 81 | ConvolutionDepthWise conv38           1 1 conv37_relu25 conv38 0=576 1=3 2=1 3=1 4=1 5=1 6=5184 7=576
 82 | ReLU             relu26           1 1 conv38 conv38_relu26
 83 | Convolution      conv39           1 1 conv38_relu26 conv39 0=96 1=1 2=1 3=1 4=0 5=1 6=55296
 84 | Eltwise          add8             2 1 add7_splitncnn_0 conv39 add8 0=1 -23301=0
 85 | Convolution      conv40           1 1 add8 conv40 0=576 1=1 2=1 3=1 4=0 5=1 6=55296
 86 | ReLU             relu27           1 1 conv40 conv40_relu27
 87 | Split            splitncnn_8      1 2 conv40_relu27 conv40_relu27_splitncnn_0 conv40_relu27_splitncnn_1
 88 | ConvolutionDepthWise conv41           1 1 conv40_relu27_splitncnn_1 conv41 0=576 1=3 2=1 3=2 4=1 5=1 6=5184 7=576
 89 | ReLU             relu28           1 1 conv41 conv41_relu28
 90 | Convolution      conv42           1 1 conv41_relu28 conv42 0=160 1=1 2=1 3=1 4=0 5=1 6=92160
 91 | Split            splitncnn_9      1 2 conv42 conv42_splitncnn_0 conv42_splitncnn_1
 92 | Convolution      conv43           1 1 conv42_splitncnn_1 conv43 0=960 1=1 2=1 3=1 4=0 5=1 6=153600
 93 | ReLU             relu29           1 1 conv43 conv43_relu29
 94 | ConvolutionDepthWise conv44           1 1 conv43_relu29 conv44 0=960 1=3 2=1 3=1 4=1 5=1 6=8640 7=960
 95 | ReLU             relu30           1 1 conv44 conv44_relu30
 96 | Convolution      conv45           1 1 conv44_relu30 conv45 0=160 1=1 2=1 3=1 4=0 5=1 6=153600
 97 | Eltwise          add9             2 1 conv42_splitncnn_0 conv45 add9 0=1 -23301=0
 98 | Split            splitncnn_10     1 2 add9 add9_splitncnn_0 add9_splitncnn_1
 99 | Convolution      conv46           1 1 add9_splitncnn_1 conv46 0=960 1=1 2=1 3=1 4=0 5=1 6=153600
100 | ReLU             relu31           1 1 conv46 conv46_relu31
101 | ConvolutionDepthWise conv47           1 1 conv46_relu31 conv47 0=960 1=3 2=1 3=1 4=1 5=1 6=8640 7=960
102 | ReLU             relu32           1 1 conv47 conv47_relu32
103 | Convolution      conv48           1 1 conv47_relu32 conv48 0=160 1=1 2=1 3=1 4=0 5=1 6=153600
104 | Eltwise          add10            2 1 add9_splitncnn_0 conv48 add10 0=1 -23301=0
105 | Convolution      conv49           1 1 add10 conv49 0=960 1=1 2=1 3=1 4=0 5=1 6=153600
106 | ReLU             relu33           1 1 conv49 conv49_relu33
107 | ConvolutionDepthWise conv50           1 1 conv49_relu33 conv50 0=960 1=3 2=1 3=1 4=1 5=1 6=8640 7=960
108 | ReLU             relu34           1 1 conv50 conv50_relu34
109 | Convolution      conv51           1 1 conv50_relu34 conv51 0=320 1=1 2=1 3=1 4=0 5=1 6=307200
110 | Convolution      conv52           1 1 conv51 conv52 0=1280 1=1 2=1 3=1 4=0 5=1 6=409600
111 | ReLU             relu35           1 1 conv52 conv52_relu35
112 | ConvolutionDepthWise yolo/conv1/dw    1 1 conv52_relu35 yolo/conv1/dw 0=1280 1=3 2=1 3=1 4=1 5=1 6=11520 7=1280
113 | ReLU             yolo/conv1/dw/relu 1 1 yolo/conv1/dw yolo/conv1/dw_yolo/conv1/dw/relu
114 | Convolution      yolo/conv1       1 1 yolo/conv1/dw_yolo/conv1/dw/relu yolo/conv1 0=576 1=1 2=1 3=1 4=0 5=1 6=737280
115 | ReLU             yolo/conv1/relu  1 1 yolo/conv1 yolo/conv1_yolo/conv1/relu
116 | Split            splitncnn_11     1 2 yolo/conv1_yolo/conv1/relu yolo/conv1_yolo/conv1/relu_splitncnn_0 yolo/conv1_yolo/conv1/relu_splitncnn_1
117 | DeconvolutionDepthWise upsample         1 1 yolo/conv1_yolo/conv1/relu_splitncnn_1 upsample 0=576 1=1 2=1 3=2 4=0 5=0 6=576 7=576
118 | Pooling          maxpool          1 1 upsample maxpool 0=0 1=2 2=1 3=1 4=0
119 | ConvolutionDepthWise yolo/conv2/dw    1 1 conv40_relu27_splitncnn_0 yolo/conv2/dw 0=576 1=3 2=1 3=1 4=1 5=1 6=5184 7=576
120 | ReLU             yolo/conv2/dw/relu 1 1 yolo/conv2/dw yolo/conv2/dw_yolo/conv2/dw/relu
121 | Convolution      yolo/conv2       1 1 yolo/conv2/dw_yolo/conv2/dw/relu yolo/conv2 0=576 1=1 2=1 3=1 4=0 5=1 6=331776
122 | ReLU             yolo/conv2/relu  1 1 yolo/conv2 yolo/conv2_yolo/conv2/relu
123 | Eltwise          yolo/conv2/sum   2 1 maxpool yolo/conv2_yolo/conv2/relu yolo/conv2/sum 0=1 -23301=0
124 | ConvolutionDepthWise yolo/conv3/dw    1 1 yolo/conv2/sum yolo/conv3/dw 0=576 1=3 2=1 3=1 4=1 5=1 6=5184 7=576
125 | ReLU             yolo/conv3/dw/relu 1 1 yolo/conv3/dw yolo/conv3/dw_yolo/conv3/dw/relu
126 | Convolution      yolo/conv3       1 1 yolo/conv3/dw_yolo/conv3/dw/relu yolo/conv3 0=576 1=1 2=1 3=1 4=0 5=1 6=331776
127 | ReLU             yolo/conv3/relu  1 1 yolo/conv3 yolo/conv3_yolo/conv3/relu
128 | Convolution      yolo/conv4       1 1 yolo/conv1_yolo/conv1/relu_splitncnn_0 yolo/conv4 0=75 1=1 2=1 3=1 4=0 5=1 6=43200
129 | Convolution      yolo/conv5       1 1 yolo/conv3_yolo/conv3/relu yolo/conv5 0=75 1=1 2=1 3=1 4=0 5=1 6=43200
130 | Yolov3DetectionOutput detection_out    2 1 yolo/conv4 yolo/conv5 detection_out 0=20 1=3 2=0.010000 3=0.450000 -23304=12,20.000000,37.000000,49.000000,94.000000,73.000000,201.000000,143.000000,265.000000,153.000000,121.000000,280.000000,279.000000 -23305=6,3.000000,4.000000,5.000000,0.000000,1.000000,2.000000 -23306=2,32.000000,16.000000 7=2
131 | 


--------------------------------------------------------------------------------