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

17
#if defined(USE_NCNN_SIMPLEOCV)
18
#include "simpleocv.h"
19
#else
20
#include <opencv2/core/core.hpp>
21
#include <opencv2/highgui/highgui.hpp>
22
#include <opencv2/imgproc/imgproc.hpp>
23
#endif
24
#include <stdio.h>
25
#include <vector>
26

27
class Noop : public ncnn::Layer
28
{
29
};
30
DEFINE_LAYER_CREATOR(Noop)
31

32
struct Object
33
{
34
    cv::Rect_<float> rect;
35
    int label;
36
    float prob;
37
};
38

39
static int detect_mobilenetv2(const cv::Mat& bgr, std::vector<Object>& objects)
40
{
41
    ncnn::Net mobilenetv2;
42

43
    mobilenetv2.opt.use_vulkan_compute = true;
44

45
    mobilenetv2.register_custom_layer("Silence", Noop_layer_creator);
46

47
    // original pretrained model from https://github.com/chuanqi305/MobileNetv2-SSDLite
48
    // https://github.com/chuanqi305/MobileNetv2-SSDLite/blob/master/ssdlite/voc/deploy.prototxt
49
    // the ncnn model https://github.com/nihui/ncnn-assets/tree/master/models
50
    if (mobilenetv2.load_param("mobilenetv2_ssdlite_voc.param"))
51
        exit(-1);
52
    if (mobilenetv2.load_model("mobilenetv2_ssdlite_voc.bin"))
53
        exit(-1);
54

55
    const int target_size = 300;
56

57
    int img_w = bgr.cols;
58
    int img_h = bgr.rows;
59

60
    ncnn::Mat in = ncnn::Mat::from_pixels_resize(bgr.data, ncnn::Mat::PIXEL_BGR, bgr.cols, bgr.rows, target_size, target_size);
61

62
    const float mean_vals[3] = {127.5f, 127.5f, 127.5f};
63
    const float norm_vals[3] = {1.0 / 127.5, 1.0 / 127.5, 1.0 / 127.5};
64
    in.substract_mean_normalize(mean_vals, norm_vals);
65

66
    ncnn::Extractor ex = mobilenetv2.create_extractor();
67

68
    ex.input("data", in);
69

70
    ncnn::Mat out;
71
    ex.extract("detection_out", out);
72

73
    //     printf("%d %d %d\n", out.w, out.h, out.c);
74
    objects.clear();
75
    for (int i = 0; i < out.h; i++)
76
    {
77
        const float* values = out.row(i);
78

79
        Object object;
80
        object.label = values[0];
81
        object.prob = values[1];
82
        object.rect.x = values[2] * img_w;
83
        object.rect.y = values[3] * img_h;
84
        object.rect.width = values[4] * img_w - object.rect.x;
85
        object.rect.height = values[5] * img_h - object.rect.y;
86

87
        objects.push_back(object);
88
    }
89

90
    return 0;
91
}
92

93
static void draw_objects(const cv::Mat& bgr, const std::vector<Object>& objects)
94
{
95
    static const char* class_names[] = {"background",
96
                                        "aeroplane", "bicycle", "bird", "boat",
97
                                        "bottle", "bus", "car", "cat", "chair",
98
                                        "cow", "diningtable", "dog", "horse",
99
                                        "motorbike", "person", "pottedplant",
100
                                        "sheep", "sofa", "train", "tvmonitor"
101
                                       };
102

103
    cv::Mat image = bgr.clone();
104

105
    for (size_t i = 0; i < objects.size(); i++)
106
    {
107
        const Object& obj = objects[i];
108

109
        fprintf(stderr, "%d = %.5f at %.2f %.2f %.2f x %.2f\n", obj.label, obj.prob,
110
                obj.rect.x, obj.rect.y, obj.rect.width, obj.rect.height);
111

112
        cv::rectangle(image, obj.rect, cv::Scalar(255, 0, 0));
113

114
        char text[256];
115
        sprintf(text, "%s %.1f%%", class_names[obj.label], obj.prob * 100);
116

117
        int baseLine = 0;
118
        cv::Size label_size = cv::getTextSize(text, cv::FONT_HERSHEY_SIMPLEX, 0.5, 1, &baseLine);
119

120
        int x = obj.rect.x;
121
        int y = obj.rect.y - label_size.height - baseLine;
122
        if (y < 0)
123
            y = 0;
124
        if (x + label_size.width > image.cols)
125
            x = image.cols - label_size.width;
126

127
        cv::rectangle(image, cv::Rect(cv::Point(x, y), cv::Size(label_size.width, label_size.height + baseLine)),
128
                      cv::Scalar(255, 255, 255), -1);
129

130
        cv::putText(image, text, cv::Point(x, y + label_size.height),
131
                    cv::FONT_HERSHEY_SIMPLEX, 0.5, cv::Scalar(0, 0, 0));
132
    }
133

134
    cv::imshow("image", image);
135
    cv::waitKey(0);
136
}
137

138
int main(int argc, char** argv)
139
{
140
    if (argc != 2)
141
    {
142
        fprintf(stderr, "Usage: %s [imagepath]\n", argv[0]);
143
        return -1;
144
    }
145

146
    const char* imagepath = argv[1];
147

148
    cv::Mat m = cv::imread(imagepath, 1);
149
    if (m.empty())
150
    {
151
        fprintf(stderr, "cv::imread %s failed\n", imagepath);
152
        return -1;
153
    }
154

155
    std::vector<Object> objects;
156
    detect_mobilenetv2(m, objects);
157

158
    draw_objects(m, objects);
159

160
    return 0;
161
}
162