1 // Tencent is pleased to support the open source community by making ncnn available.
2 //
3 // Copyright (C) 2018 THL A29 Limited, a Tencent company. All rights reserved.
4 //
5 // Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
6 // in compliance with the License. You may obtain a copy of the License at
7 //
8 // https://opensource.org/licenses/BSD-3-Clause
9 //
10 // Unless required by applicable law or agreed to in writing, software distributed
11 // under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
12 // CONDITIONS OF ANY KIND, either express or implied. See the License for the
13 // specific language governing permissions and limitations under the License.
14
15 #include "net.h"
16 #include "platform.h"
17
18 #if defined(USE_NCNN_SIMPLEOCV)
19 #include "simpleocv.h"
20 #else
21 #include <opencv2/core/core.hpp>
22 #include <opencv2/highgui/highgui.hpp>
23 #include <opencv2/imgproc/imgproc.hpp>
24 #endif
25 #include <stdio.h>
26 #include <vector>
27 #if NCNN_VULKAN
28 #include "gpu.h"
29 #endif // NCNN_VULKAN
30
31 template<class T>
clamp(const T & v,const T & lo,const T & hi)32 const T& clamp(const T& v, const T& lo, const T& hi)
33 {
34 assert(!(hi < lo));
35 return v < lo ? lo : hi < v ? hi : v;
36 }
37
38 struct Object
39 {
40 cv::Rect_<float> rect;
41 int label;
42 float prob;
43 };
44
detect_mobilenetv3(const cv::Mat & bgr,std::vector<Object> & objects)45 static int detect_mobilenetv3(const cv::Mat& bgr, std::vector<Object>& objects)
46 {
47 ncnn::Net mobilenetv3;
48
49 #if NCNN_VULKAN
50 mobilenetv3.opt.use_vulkan_compute = true;
51 #endif // NCNN_VULKAN
52
53 // converted ncnn model from https://github.com/ujsyehao/mobilenetv3-ssd
54 mobilenetv3.load_param("./mobilenetv3_ssdlite_voc.param");
55 mobilenetv3.load_model("./mobilenetv3_ssdlite_voc.bin");
56
57 const int target_size = 300;
58
59 int img_w = bgr.cols;
60 int img_h = bgr.rows;
61
62 ncnn::Mat in = ncnn::Mat::from_pixels_resize(bgr.data, ncnn::Mat::PIXEL_BGR2RGB, bgr.cols, bgr.rows, target_size, target_size);
63
64 const float mean_vals[3] = {123.675f, 116.28f, 103.53f};
65 const float norm_vals[3] = {1.0f, 1.0f, 1.0f};
66 in.substract_mean_normalize(mean_vals, norm_vals);
67
68 ncnn::Extractor ex = mobilenetv3.create_extractor();
69
70 ex.input("input", in);
71
72 ncnn::Mat out;
73 ex.extract("detection_out", out);
74
75 // printf("%d %d %d\n", out.w, out.h, out.c);
76 objects.clear();
77 for (int i = 0; i < out.h; i++)
78 {
79 const float* values = out.row(i);
80
81 Object object;
82 object.label = values[0];
83 object.prob = values[1];
84
85 // filter out cross-boundary
86 float x1 = clamp(values[2] * target_size, 0.f, float(target_size - 1)) / target_size * img_w;
87 float y1 = clamp(values[3] * target_size, 0.f, float(target_size - 1)) / target_size * img_h;
88 float x2 = clamp(values[4] * target_size, 0.f, float(target_size - 1)) / target_size * img_w;
89 float y2 = clamp(values[5] * target_size, 0.f, float(target_size - 1)) / target_size * img_h;
90
91 object.rect.x = x1;
92 object.rect.y = y1;
93 object.rect.width = x2 - x1;
94 object.rect.height = y2 - y1;
95
96 objects.push_back(object);
97 }
98
99 return 0;
100 }
101
draw_objects(const cv::Mat & bgr,const std::vector<Object> & objects)102 static void draw_objects(const cv::Mat& bgr, const std::vector<Object>& objects)
103 {
104 static const char* class_names[] = {"background",
105 "aeroplane", "bicycle", "bird", "boat",
106 "bottle", "bus", "car", "cat", "chair",
107 "cow", "diningtable", "dog", "horse",
108 "motorbike", "person", "pottedplant",
109 "sheep", "sofa", "train", "tvmonitor"
110 };
111
112 cv::Mat image = bgr.clone();
113
114 for (size_t i = 0; i < objects.size(); i++)
115 {
116 if (objects[i].prob > 0.6)
117 {
118 const Object& obj = objects[i];
119
120 fprintf(stderr, "%d = %.5f at %.2f %.2f %.2f x %.2f\n", obj.label, obj.prob,
121 obj.rect.x, obj.rect.y, obj.rect.width, obj.rect.height);
122
123 cv::rectangle(image, obj.rect, cv::Scalar(255, 0, 0));
124
125 char text[256];
126 sprintf(text, "%s %.1f%%", class_names[obj.label], obj.prob * 100);
127
128 int baseLine = 0;
129 cv::Size label_size = cv::getTextSize(text, cv::FONT_HERSHEY_SIMPLEX, 0.5, 1, &baseLine);
130
131 int x = obj.rect.x;
132 int y = obj.rect.y - label_size.height - baseLine;
133 if (y < 0)
134 y = 0;
135 if (x + label_size.width > image.cols)
136 x = image.cols - label_size.width;
137
138 cv::rectangle(image, cv::Rect(cv::Point(x, y), cv::Size(label_size.width, label_size.height + baseLine)),
139 cv::Scalar(255, 255, 255), -1);
140
141 cv::putText(image, text, cv::Point(x, y + label_size.height),
142 cv::FONT_HERSHEY_SIMPLEX, 0.5, cv::Scalar(0, 0, 0));
143 }
144 }
145
146 cv::imshow("image", image);
147 cv::waitKey(0);
148 }
149
main(int argc,char ** argv)150 int main(int argc, char** argv)
151 {
152 if (argc != 2)
153 {
154 fprintf(stderr, "Usage: %s [imagepath]\n", argv[0]);
155 return -1;
156 }
157
158 const char* imagepath = argv[1];
159
160 cv::Mat m = cv::imread(imagepath, 1);
161 if (m.empty())
162 {
163 fprintf(stderr, "cv::imread %s failed\n", imagepath);
164 return -1;
165 }
166
167 std::vector<Object> objects;
168 detect_mobilenetv3(m, objects);
169
170 draw_objects(m, objects);
171
172 return 0;
173 }
174