ncnn

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>
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

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
    if (mobilenetv3.load_param("./mobilenetv3_ssdlite_voc.param"))
55
        exit(-1);
56
    if (mobilenetv3.load_model("./mobilenetv3_ssdlite_voc.bin"))
57
        exit(-1);
58

59
    const int target_size = 300;
60

61
    int img_w = bgr.cols;
62
    int img_h = bgr.rows;
63

64
    ncnn::Mat in = ncnn::Mat::from_pixels_resize(bgr.data, ncnn::Mat::PIXEL_BGR2RGB, bgr.cols, bgr.rows, target_size, target_size);
65

66
    const float mean_vals[3] = {123.675f, 116.28f, 103.53f};
67
    const float norm_vals[3] = {1.0f, 1.0f, 1.0f};
68
    in.substract_mean_normalize(mean_vals, norm_vals);
69

70
    ncnn::Extractor ex = mobilenetv3.create_extractor();
71

72
    ex.input("input", in);
73

74
    ncnn::Mat out;
75
    ex.extract("detection_out", out);
76

77
    //     printf("%d %d %d\n", out.w, out.h, out.c);
78
    objects.clear();
79
    for (int i = 0; i < out.h; i++)
80
    {
81
        const float* values = out.row(i);
82

83
        Object object;
84
        object.label = values[0];
85
        object.prob = values[1];
86

87
        // filter out cross-boundary
88
        float x1 = clamp(values[2] * target_size, 0.f, float(target_size - 1)) / target_size * img_w;
89
        float y1 = clamp(values[3] * target_size, 0.f, float(target_size - 1)) / target_size * img_h;
90
        float x2 = clamp(values[4] * target_size, 0.f, float(target_size - 1)) / target_size * img_w;
91
        float y2 = clamp(values[5] * target_size, 0.f, float(target_size - 1)) / target_size * img_h;
92

93
        object.rect.x = x1;
94
        object.rect.y = y1;
95
        object.rect.width = x2 - x1;
96
        object.rect.height = y2 - y1;
97

98
        objects.push_back(object);
99
    }
100

101
    return 0;
102
}
103

104
static void draw_objects(const cv::Mat& bgr, const std::vector<Object>& objects)
105
{
106
    static const char* class_names[] = {"background",
107
                                        "aeroplane", "bicycle", "bird", "boat",
108
                                        "bottle", "bus", "car", "cat", "chair",
109
                                        "cow", "diningtable", "dog", "horse",
110
                                        "motorbike", "person", "pottedplant",
111
                                        "sheep", "sofa", "train", "tvmonitor"
112
                                       };
113

114
    cv::Mat image = bgr.clone();
115

116
    for (size_t i = 0; i < objects.size(); i++)
117
    {
118
        if (objects[i].prob > 0.6)
119
        {
120
            const Object& obj = objects[i];
121

122
            fprintf(stderr, "%d = %.5f at %.2f %.2f %.2f x %.2f\n", obj.label, obj.prob,
123
                    obj.rect.x, obj.rect.y, obj.rect.width, obj.rect.height);
124

125
            cv::rectangle(image, obj.rect, cv::Scalar(255, 0, 0));
126

127
            char text[256];
128
            sprintf(text, "%s %.1f%%", class_names[obj.label], obj.prob * 100);
129

130
            int baseLine = 0;
131
            cv::Size label_size = cv::getTextSize(text, cv::FONT_HERSHEY_SIMPLEX, 0.5, 1, &baseLine);
132

133
            int x = obj.rect.x;
134
            int y = obj.rect.y - label_size.height - baseLine;
135
            if (y < 0)
136
                y = 0;
137
            if (x + label_size.width > image.cols)
138
                x = image.cols - label_size.width;
139

140
            cv::rectangle(image, cv::Rect(cv::Point(x, y), cv::Size(label_size.width, label_size.height + baseLine)),
141
                          cv::Scalar(255, 255, 255), -1);
142

143
            cv::putText(image, text, cv::Point(x, y + label_size.height),
144
                        cv::FONT_HERSHEY_SIMPLEX, 0.5, cv::Scalar(0, 0, 0));
145
        }
146
    }
147

148
    cv::imshow("image", image);
149
    cv::waitKey(0);
150
}
151

152
int main(int argc, char** argv)
153
{
154
    if (argc != 2)
155
    {
156
        fprintf(stderr, "Usage: %s [imagepath]\n", argv[0]);
157
        return -1;
158
    }
159

160
    const char* imagepath = argv[1];
161

162
    cv::Mat m = cv::imread(imagepath, 1);
163
    if (m.empty())
164
    {
165
        fprintf(stderr, "cv::imread %s failed\n", imagepath);
166
        return -1;
167
    }
168

169
    std::vector<Object> objects;
170
    detect_mobilenetv3(m, objects);
171

172
    draw_objects(m, objects);
173

174
    return 0;
175
}
176