1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
|
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
#include "test_precomp.hpp"
namespace opencv_test { namespace {
// label format:
// image_name
// num_face
// face_1
// face_..
// face_num
std::map<std::string, Mat> blobFromTXT(const std::string& path, int numCoords)
{
std::ifstream ifs(path.c_str());
CV_Assert(ifs.is_open());
std::map<std::string, Mat> gt;
Mat faces;
int faceNum = -1;
int faceCount = 0;
for (std::string line, key; getline(ifs, line); )
{
std::istringstream iss(line);
if (line.find(".png") != std::string::npos)
{
// Get filename
iss >> key;
}
else if (line.find(" ") == std::string::npos)
{
// Get the number of faces
iss >> faceNum;
}
else
{
// Get faces
Mat face(1, numCoords, CV_32FC1);
for (int j = 0; j < numCoords; j++)
{
iss >> face.at<float>(0, j);
}
faces.push_back(face);
faceCount++;
}
if (faceCount == faceNum)
{
// Store faces
gt[key] = faces;
faces.release();
faceNum = -1;
faceCount = 0;
}
}
return gt;
}
TEST(Objdetect_face_detection, regression)
{
// Pre-set params
float scoreThreshold = 0.7f;
float matchThreshold = 0.7f;
float l2disThreshold = 15.0f;
int numLM = 5;
int numCoords = 4 + 2 * numLM;
// Load ground truth labels
std::map<std::string, Mat> gt = blobFromTXT(findDataFile("dnn_face/detection/cascades_labels.txt"), numCoords);
// Initialize detector
std::string model = findDataFile("dnn/onnx/models/yunet-202303.onnx", false);
Ptr<FaceDetectorYN> faceDetector = FaceDetectorYN::create(model, "", Size(300, 300));
faceDetector->setScoreThreshold(0.7f);
// Detect and match
for (auto item: gt)
{
std::string imagePath = findDataFile("cascadeandhog/images/" + item.first);
Mat image = imread(imagePath);
// Set input size
faceDetector->setInputSize(image.size());
// Run detection
Mat faces;
faceDetector->detect(image, faces);
// std::cout << item.first << " " << item.second.rows << " " << faces.rows << std::endl;
// Match bboxes and landmarks
std::vector<bool> matchedItem(item.second.rows, false);
for (int i = 0; i < faces.rows; i++)
{
if (faces.at<float>(i, numCoords) < scoreThreshold)
continue;
bool boxMatched = false;
std::vector<bool> lmMatched(numLM, false);
cv::Rect2f resBox(faces.at<float>(i, 0), faces.at<float>(i, 1), faces.at<float>(i, 2), faces.at<float>(i, 3));
for (int j = 0; j < item.second.rows && !boxMatched; j++)
{
if (matchedItem[j])
continue;
// Retrieve bbox and compare IoU
cv::Rect2f gtBox(item.second.at<float>(j, 0), item.second.at<float>(j, 1), item.second.at<float>(j, 2), item.second.at<float>(j, 3));
double interArea = (resBox & gtBox).area();
double iou = interArea / (resBox.area() + gtBox.area() - interArea);
if (iou >= matchThreshold)
{
boxMatched = true;
matchedItem[j] = true;
}
// Match landmarks if bbox is matched
if (!boxMatched)
continue;
for (int lmIdx = 0; lmIdx < numLM; lmIdx++)
{
float gtX = item.second.at<float>(j, 4 + 2 * lmIdx);
float gtY = item.second.at<float>(j, 4 + 2 * lmIdx + 1);
float resX = faces.at<float>(i, 4 + 2 * lmIdx);
float resY = faces.at<float>(i, 4 + 2 * lmIdx + 1);
float l2dis = cv::sqrt((gtX - resX) * (gtX - resX) + (gtY - resY) * (gtY - resY));
if (l2dis <= l2disThreshold)
{
lmMatched[lmIdx] = true;
}
}
break;
}
EXPECT_TRUE(boxMatched) << "In image " << item.first << ", cannot match resBox " << resBox << " with any ground truth.";
if (boxMatched)
{
EXPECT_TRUE(std::all_of(lmMatched.begin(), lmMatched.end(), [](bool v) { return v; })) << "In image " << item.first << ", resBox " << resBox << " matched but its landmarks failed to match.";
}
}
}
}
TEST(Objdetect_face_recognition, regression)
{
// Pre-set params
float score_thresh = 0.9f;
float nms_thresh = 0.3f;
double cosine_similar_thresh = 0.363;
double l2norm_similar_thresh = 1.128;
// Load ground truth labels
std::ifstream ifs(findDataFile("dnn_face/recognition/cascades_label.txt").c_str());
CV_Assert(ifs.is_open());
std::set<std::string> fSet;
std::map<std::string, Mat> featureMap;
std::map<std::pair<std::string, std::string>, int> gtMap;
for (std::string line, key; getline(ifs, line);)
{
std::string fname1, fname2;
int label;
std::istringstream iss(line);
iss>>fname1>>fname2>>label;
// std::cout<<fname1<<" "<<fname2<<" "<<label<<std::endl;
fSet.insert(fname1);
fSet.insert(fname2);
gtMap[std::make_pair(fname1, fname2)] = label;
}
// Initialize detector
std::string detect_model = findDataFile("dnn/onnx/models/yunet-202303.onnx", false);
Ptr<FaceDetectorYN> faceDetector = FaceDetectorYN::create(detect_model, "", Size(150, 150), score_thresh, nms_thresh);
std::string recog_model = findDataFile("dnn/onnx/models/face_recognizer_fast.onnx", false);
Ptr<FaceRecognizerSF> faceRecognizer = FaceRecognizerSF::create(recog_model, "");
// Detect and match
for (auto fname: fSet)
{
std::string imagePath = findDataFile("dnn_face/recognition/" + fname);
Mat image = imread(imagePath);
Mat faces;
faceDetector->detect(image, faces);
Mat aligned_face;
faceRecognizer->alignCrop(image, faces.row(0), aligned_face);
Mat feature;
faceRecognizer->feature(aligned_face, feature);
featureMap[fname] = feature.clone();
}
for (auto item: gtMap)
{
Mat feature1 = featureMap[item.first.first];
Mat feature2 = featureMap[item.first.second];
int label = item.second;
double cos_score = faceRecognizer->match(feature1, feature2, FaceRecognizerSF::DisType::FR_COSINE);
double L2_score = faceRecognizer->match(feature1, feature2, FaceRecognizerSF::DisType::FR_NORM_L2);
EXPECT_TRUE(label == 0 ? cos_score <= cosine_similar_thresh : cos_score > cosine_similar_thresh) << "Cosine match result of images " << item.first.first << " and " << item.first.second << " is different from ground truth (score: "<< cos_score <<";Thresh: "<< cosine_similar_thresh <<").";
EXPECT_TRUE(label == 0 ? L2_score > l2norm_similar_thresh : L2_score <= l2norm_similar_thresh) << "L2norm match result of images " << item.first.first << " and " << item.first.second << " is different from ground truth (score: "<< L2_score <<";Thresh: "<< l2norm_similar_thresh <<").";
}
}
}} // namespace
|
