inputs = net.getInputsInfo();
outputs = net.getOutputsInfo();
layers.resize(net.layerCount()); // A hack to execute InfEngineBackendNet::layerCount correctly.
- initPlugin(net);
}
void InfEngineBackendNet::Release() noexcept
Mat out = net.forward(outputLayer).clone();
if (outputLayer == "detection_out")
- checkDetections(outDefault, out, "First run", l1, lInf);
+ normAssertDetections(outDefault, out, "First run", 0.2);
else
normAssert(outDefault, out, "First run", l1, lInf);
Mat out = net.forward("detection_out");
Mat ref = blobFromNPY(_tf("ssd_out.npy"));
- normAssert(ref, out);
+ normAssertDetections(ref, out);
}
typedef testing::TestWithParam<DNNTarget> Reproducibility_MobileNet_SSD;
Mat out = net.forward();
Mat ref = blobFromNPY(_tf("mobilenet_ssd_caffe_out.npy"));
- normAssert(ref, out);
+ normAssertDetections(ref, out);
// Check that detections aren't preserved.
inp.setTo(0.0f);
// Output has shape 1x1xNx7 where N - number of detections.
// An every detection is a vector of values [id, classId, confidence, left, top, right, bottom]
Mat out = net.forward();
-
- Mat ref = (Mat_<float>(6, 5) << 0.99520785, 0.80997437, 0.16379407, 0.87996572, 0.26685631,
- 0.9934696, 0.2831718, 0.50738752, 0.345781, 0.5985168,
- 0.99096733, 0.13629119, 0.24892329, 0.19756334, 0.3310290,
- 0.98977017, 0.23901358, 0.09084064, 0.29902688, 0.1769477,
- 0.97203469, 0.67965847, 0.06876482, 0.73999709, 0.1513494,
- 0.95097077, 0.51901293, 0.45863652, 0.5777427, 0.5347801);
- normAssert(out.reshape(1, out.total() / 7).rowRange(0, 6).colRange(2, 7), ref);
+ Mat ref = (Mat_<float>(6, 7) << 0, 1, 0.99520785, 0.80997437, 0.16379407, 0.87996572, 0.26685631,
+ 0, 1, 0.9934696, 0.2831718, 0.50738752, 0.345781, 0.5985168,
+ 0, 1, 0.99096733, 0.13629119, 0.24892329, 0.19756334, 0.3310290,
+ 0, 1, 0.98977017, 0.23901358, 0.09084064, 0.29902688, 0.1769477,
+ 0, 1, 0.97203469, 0.67965847, 0.06876482, 0.73999709, 0.1513494,
+ 0, 1, 0.95097077, 0.51901293, 0.45863652, 0.5777427, 0.5347801);
+ normAssertDetections(ref, out, "", 0.5, 1e-5, 2e-4);
}
INSTANTIATE_TEST_CASE_P(Test_Caffe, opencv_face_detector,
Combine(
"resnet50_rfcn_final.caffemodel"};
std::string protos[] = {"faster_rcnn_vgg16.prototxt", "faster_rcnn_zf.prototxt",
"rfcn_pascal_voc_resnet50.prototxt"};
- Mat refs[] = {(Mat_<float>(3, 6) << 2, 0.949398, 99.2454, 210.141, 601.205, 462.849,
- 7, 0.997022, 481.841, 92.3218, 722.685, 175.953,
- 12, 0.993028, 133.221, 189.377, 350.994, 563.166),
- (Mat_<float>(3, 6) << 2, 0.90121, 120.407, 115.83, 570.586, 528.395,
- 7, 0.988779, 469.849, 75.1756, 718.64, 186.762,
- 12, 0.967198, 138.588, 206.843, 329.766, 553.176),
- (Mat_<float>(2, 6) << 7, 0.991359, 491.822, 81.1668, 702.573, 178.234,
- 12, 0.94786, 132.093, 223.903, 338.077, 566.16)};
+ Mat refs[] = {(Mat_<float>(3, 7) << 0, 2, 0.949398, 99.2454, 210.141, 601.205, 462.849,
+ 0, 7, 0.997022, 481.841, 92.3218, 722.685, 175.953,
+ 0, 12, 0.993028, 133.221, 189.377, 350.994, 563.166),
+ (Mat_<float>(3, 7) << 0, 2, 0.90121, 120.407, 115.83, 570.586, 528.395,
+ 0, 7, 0.988779, 469.849, 75.1756, 718.64, 186.762,
+ 0, 12, 0.967198, 138.588, 206.843, 329.766, 553.176),
+ (Mat_<float>(2, 7) << 0, 7, 0.991359, 491.822, 81.1668, 702.573, 178.234,
+ 0, 12, 0.94786, 132.093, 223.903, 338.077, 566.16)};
for (int i = 0; i < 3; ++i)
{
std::string proto = findDataFile("dnn/" + protos[i], false);
// Output has shape 1x1xNx7 where N - number of detections.
// An every detection is a vector of values [id, classId, confidence, left, top, right, bottom]
Mat out = net.forward();
- out = out.reshape(1, out.total() / 7);
-
- Mat detections;
- for (int j = 0; j < out.rows; ++j)
- {
- if (out.at<float>(j, 2) > 0.8)
- detections.push_back(out.row(j).colRange(1, 7));
- }
- normAssert(detections, refs[i], ("model name: " + models[i]).c_str(), 1e-3, 1e-3);
+ normAssertDetections(refs[i], out, ("model name: " + models[i]).c_str(), 0.8);
}
}
EXPECT_LE(normInf, lInf) << comment;
}
+static std::vector<cv::Rect2d> matToBoxes(const cv::Mat& m)
+{
+ EXPECT_EQ(m.type(), CV_32FC1);
+ EXPECT_EQ(m.dims, 2);
+ EXPECT_EQ(m.cols, 4);
+
+ std::vector<cv::Rect2d> boxes(m.rows);
+ for (int i = 0; i < m.rows; ++i)
+ {
+ CV_Assert(m.row(i).isContinuous());
+ const float* data = m.ptr<float>(i);
+ double l = data[0], t = data[1], r = data[2], b = data[3];
+ boxes[i] = cv::Rect2d(l, t, r - l, b - t);
+ }
+ return boxes;
+}
+
+inline void normAssertDetections(const std::vector<int>& refClassIds,
+ const std::vector<float>& refScores,
+ const std::vector<cv::Rect2d>& refBoxes,
+ const std::vector<int>& testClassIds,
+ const std::vector<float>& testScores,
+ const std::vector<cv::Rect2d>& testBoxes,
+ const char *comment = "", double confThreshold = 0.0,
+ double scores_diff = 1e-5, double boxes_iou_diff = 1e-4)
+{
+ std::vector<bool> matchedRefBoxes(refBoxes.size(), false);
+ for (int i = 0; i < testBoxes.size(); ++i)
+ {
+ double testScore = testScores[i];
+ if (testScore < confThreshold)
+ continue;
+
+ int testClassId = testClassIds[i];
+ const cv::Rect2d& testBox = testBoxes[i];
+ bool matched = false;
+ for (int j = 0; j < refBoxes.size() && !matched; ++j)
+ {
+ if (!matchedRefBoxes[j] && testClassId == refClassIds[j] &&
+ std::abs(testScore - refScores[j]) < scores_diff)
+ {
+ double interArea = (testBox & refBoxes[j]).area();
+ double iou = interArea / (testBox.area() + refBoxes[j].area() - interArea);
+ if (std::abs(iou - 1.0) < boxes_iou_diff)
+ {
+ matched = true;
+ matchedRefBoxes[j] = true;
+ }
+ }
+ }
+ if (!matched)
+ std::cout << cv::format("Unmatched prediction: class %d score %f box ",
+ testClassId, testScore) << testBox << std::endl;
+ EXPECT_TRUE(matched) << comment;
+ }
+
+ // Check unmatched reference detections.
+ for (int i = 0; i < refBoxes.size(); ++i)
+ {
+ if (!matchedRefBoxes[i] && refScores[i] > confThreshold)
+ {
+ std::cout << cv::format("Unmatched reference: class %d score %f box ",
+ refClassIds[i], refScores[i]) << refBoxes[i] << std::endl;
+ EXPECT_LE(refScores[i], confThreshold) << comment;
+ }
+ }
+}
+
+// For SSD-based object detection networks which produce output of shape 1x1xNx7
+// where N is a number of detections and an every detection is represented by
+// a vector [batchId, classId, confidence, left, top, right, bottom].
+inline void normAssertDetections(cv::Mat ref, cv::Mat out, const char *comment = "",
+ double confThreshold = 0.0, double scores_diff = 1e-5,
+ double boxes_iou_diff = 1e-4)
+{
+ CV_Assert(ref.total() % 7 == 0);
+ CV_Assert(out.total() % 7 == 0);
+ ref = ref.reshape(1, ref.total() / 7);
+ out = out.reshape(1, out.total() / 7);
+
+ cv::Mat refClassIds, testClassIds;
+ ref.col(1).convertTo(refClassIds, CV_32SC1);
+ out.col(1).convertTo(testClassIds, CV_32SC1);
+ std::vector<float> refScores(ref.col(2)), testScores(out.col(2));
+ std::vector<cv::Rect2d> refBoxes = matToBoxes(ref.colRange(3, 7));
+ std::vector<cv::Rect2d> testBoxes = matToBoxes(out.colRange(3, 7));
+ normAssertDetections(refClassIds, refScores, refBoxes, testClassIds, testScores,
+ testBoxes, comment, confThreshold, scores_diff, boxes_iou_diff);
+}
+
inline bool readFileInMemory(const std::string& filename, std::string& content)
{
std::ios::openmode mode = std::ios::in | std::ios::binary;
const std::vector<cv::String>& outNames,
const std::vector<int>& refClassIds,
const std::vector<float>& refConfidences,
- const std::vector<Rect2f>& refBoxes,
+ const std::vector<Rect2d>& refBoxes,
int targetId, float confThreshold = 0.24)
{
Mat sample = imread(_tf("dog416.png"));
std::vector<int> classIds;
std::vector<float> confidences;
- std::vector<Rect2f> boxes;
+ std::vector<Rect2d> boxes;
for (int i = 0; i < outs.size(); ++i)
{
Mat& out = outs[i];
double confidence;
Point maxLoc;
minMaxLoc(scores, 0, &confidence, 0, &maxLoc);
- if (confidence > confThreshold)
- {
- float* detection = out.ptr<float>(j);
- float centerX = detection[0];
- float centerY = detection[1];
- float width = detection[2];
- float height = detection[3];
- boxes.push_back(Rect2f(centerX - 0.5 * width, centerY - 0.5 * height,
- width, height));
- confidences.push_back(confidence);
- classIds.push_back(maxLoc.x);
- }
- }
- }
- ASSERT_EQ(classIds.size(), refClassIds.size());
- ASSERT_EQ(confidences.size(), refConfidences.size());
- ASSERT_EQ(boxes.size(), refBoxes.size());
- for (int i = 0; i < boxes.size(); ++i)
- {
- ASSERT_EQ(classIds[i], refClassIds[i]);
- ASSERT_LE(std::abs(confidences[i] - refConfidences[i]), 1e-4);
- float iou = (boxes[i] & refBoxes[i]).area() / (boxes[i] | refBoxes[i]).area();
- ASSERT_LE(std::abs(iou - 1.0f), 1e-4);
+ float* detection = out.ptr<float>(j);
+ double centerX = detection[0];
+ double centerY = detection[1];
+ double width = detection[2];
+ double height = detection[3];
+ boxes.push_back(Rect2d(centerX - 0.5 * width, centerY - 0.5 * height,
+ width, height));
+ confidences.push_back(confidence);
+ classIds.push_back(maxLoc.x);
+ }
}
+ normAssertDetections(refClassIds, refConfidences, refBoxes, classIds,
+ confidences, boxes, "", confThreshold, 8e-5, 3e-5);
}
typedef testing::TestWithParam<DNNTarget> Test_Darknet_nets;
std::vector<int> classIds(3);
std::vector<float> confidences(3);
- std::vector<Rect2f> boxes(3);
- classIds[0] = 6; confidences[0] = 0.750469f; boxes[0] = Rect2f(0.577374, 0.127391, 0.325575, 0.173418); // a car
- classIds[1] = 1; confidences[1] = 0.780879f; boxes[1] = Rect2f(0.270762, 0.264102, 0.461713, 0.48131); // a bycicle
- classIds[2] = 11; confidences[2] = 0.901615f; boxes[2] = Rect2f(0.1386, 0.338509, 0.282737, 0.60028); // a dog
+ std::vector<Rect2d> boxes(3);
+ classIds[0] = 6; confidences[0] = 0.750469f; boxes[0] = Rect2d(0.577374, 0.127391, 0.325575, 0.173418); // a car
+ classIds[1] = 1; confidences[1] = 0.780879f; boxes[1] = Rect2d(0.270762, 0.264102, 0.461713, 0.48131); // a bycicle
+ classIds[2] = 11; confidences[2] = 0.901615f; boxes[2] = Rect2d(0.1386, 0.338509, 0.282737, 0.60028); // a dog
testDarknetModel("yolo-voc.cfg", "yolo-voc.weights", outNames,
classIds, confidences, boxes, targetId);
}
std::vector<cv::String> outNames(1, "detection_out");
std::vector<int> classIds(2);
std::vector<float> confidences(2);
- std::vector<Rect2f> boxes(2);
- classIds[0] = 6; confidences[0] = 0.761967f; boxes[0] = Rect2f(0.579042, 0.159161, 0.31544, 0.160779); // a car
- classIds[1] = 11; confidences[1] = 0.780595f; boxes[1] = Rect2f(0.129696, 0.386467, 0.315579, 0.534527); // a dog
+ std::vector<Rect2d> boxes(2);
+ classIds[0] = 6; confidences[0] = 0.761967f; boxes[0] = Rect2d(0.579042, 0.159161, 0.31544, 0.160779); // a car
+ classIds[1] = 11; confidences[1] = 0.780595f; boxes[1] = Rect2d(0.129696, 0.386467, 0.315579, 0.534527); // a dog
testDarknetModel("tiny-yolo-voc.cfg", "tiny-yolo-voc.weights", outNames,
classIds, confidences, boxes, targetId);
}
std::vector<int> classIds(3);
std::vector<float> confidences(3);
- std::vector<Rect2f> boxes(3);
- classIds[0] = 7; confidences[0] = 0.952983f; boxes[0] = Rect2f(0.614622, 0.150257, 0.286747, 0.138994); // a truck
- classIds[1] = 1; confidences[1] = 0.987908f; boxes[1] = Rect2f(0.150913, 0.221933, 0.591342, 0.524327); // a bycicle
- classIds[2] = 16; confidences[2] = 0.998836f; boxes[2] = Rect2f(0.160024, 0.389964, 0.257861, 0.553752); // a dog (COCO)
+ std::vector<Rect2d> boxes(3);
+ classIds[0] = 7; confidences[0] = 0.952983f; boxes[0] = Rect2d(0.614622, 0.150257, 0.286747, 0.138994); // a truck
+ classIds[1] = 1; confidences[1] = 0.987908f; boxes[1] = Rect2d(0.150913, 0.221933, 0.591342, 0.524327); // a bycicle
+ classIds[2] = 16; confidences[2] = 0.998836f; boxes[2] = Rect2d(0.160024, 0.389964, 0.257861, 0.553752); // a dog (COCO)
testDarknetModel("yolov3.cfg", "yolov3.weights", outNames,
classIds, confidences, boxes, targetId);
}
normAssert(target[0].reshape(1, 1), output[0].reshape(1, 1), "", 1e-5, 1.5e-4);
normAssert(target[1].reshape(1, 1), output[1].reshape(1, 1), "", 1e-5, 3e-4);
- normAssert(target[2].reshape(1, 1), output[2].reshape(1, 1), "", 4e-5, 1e-2);
+ normAssertDetections(target[2], output[2], "", 0.2);
}
TEST_P(Test_TensorFlow_nets, Inception_v2_SSD)
// Output has shape 1x1xNx7 where N - number of detections.
// An every detection is a vector of values [id, classId, confidence, left, top, right, bottom]
Mat out = net.forward();
- out = out.reshape(1, out.total() / 7);
-
- Mat detections;
- for (int i = 0; i < out.rows; ++i)
- {
- if (out.at<float>(i, 2) > 0.5)
- detections.push_back(out.row(i).colRange(1, 7));
- }
-
- Mat ref = (Mat_<float>(5, 6) << 1, 0.90176028, 0.19872092, 0.36311883, 0.26461923, 0.63498729,
- 3, 0.93569964, 0.64865261, 0.45906419, 0.80675775, 0.65708131,
- 3, 0.75838411, 0.44668293, 0.45907149, 0.49459291, 0.52197015,
- 10, 0.95932811, 0.38349164, 0.32528657, 0.40387636, 0.39165527,
- 10, 0.93973452, 0.66561931, 0.37841269, 0.68074018, 0.42907384);
- normAssert(detections, ref);
+ Mat ref = (Mat_<float>(5, 7) << 0, 1, 0.90176028, 0.19872092, 0.36311883, 0.26461923, 0.63498729,
+ 0, 3, 0.93569964, 0.64865261, 0.45906419, 0.80675775, 0.65708131,
+ 0, 3, 0.75838411, 0.44668293, 0.45907149, 0.49459291, 0.52197015,
+ 0, 10, 0.95932811, 0.38349164, 0.32528657, 0.40387636, 0.39165527,
+ 0, 10, 0.93973452, 0.66561931, 0.37841269, 0.68074018, 0.42907384);
+ normAssertDetections(ref, out, "", 0.5);
}
TEST_P(Test_TensorFlow_nets, opencv_face_detector_uint8)
Mat out = net.forward();
// References are from test for Caffe model.
- Mat ref = (Mat_<float>(6, 5) << 0.99520785, 0.80997437, 0.16379407, 0.87996572, 0.26685631,
- 0.9934696, 0.2831718, 0.50738752, 0.345781, 0.5985168,
- 0.99096733, 0.13629119, 0.24892329, 0.19756334, 0.3310290,
- 0.98977017, 0.23901358, 0.09084064, 0.29902688, 0.1769477,
- 0.97203469, 0.67965847, 0.06876482, 0.73999709, 0.1513494,
- 0.95097077, 0.51901293, 0.45863652, 0.5777427, 0.5347801);
- normAssert(out.reshape(1, out.total() / 7).rowRange(0, 6).colRange(2, 7), ref, "", 2.8e-4, 3.4e-3);
+ Mat ref = (Mat_<float>(6, 7) << 0, 1, 0.99520785, 0.80997437, 0.16379407, 0.87996572, 0.26685631,
+ 0, 1, 0.9934696, 0.2831718, 0.50738752, 0.345781, 0.5985168,
+ 0, 1, 0.99096733, 0.13629119, 0.24892329, 0.19756334, 0.3310290,
+ 0, 1, 0.98977017, 0.23901358, 0.09084064, 0.29902688, 0.1769477,
+ 0, 1, 0.97203469, 0.67965847, 0.06876482, 0.73999709, 0.1513494,
+ 0, 1, 0.95097077, 0.51901293, 0.45863652, 0.5777427, 0.5347801);
+ normAssertDetections(ref, out, "", 0.9, 3.4e-3, 1e-2);
}
INSTANTIATE_TEST_CASE_P(/**/, Test_TensorFlow_nets, availableDnnTargets());