mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2025-01-18 14:13:15 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge pull request #17743 from alalek:test_17666

Browse Source
This commit is contained in:
Alexander Alekhin 2020-07-06 22:36:35 +00:00
commit d62e0a3695
4 changed files with 91 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
modules/dnn/perf/perf_net.cpp
View File

@ -216,6 +216,19 @@ PERF_TEST_P_(DNNTestNetwork, YOLOv4)
processNet("dnn/yolov4.weights", "dnn/yolov4.cfg", "", inp);
}
PERF_TEST_P_(DNNTestNetwork, YOLOv4_tiny)
{
if (backend == DNN_BACKEND_HALIDE)
throw SkipTestException("");
if (target == DNN_TARGET_MYRIAD)
throw SkipTestException("");
Mat sample = imread(findDataFile("dnn/dog416.png"));
cvtColor(sample, sample, COLOR_BGR2RGB);
Mat inp;
sample.convertTo(inp, CV_32FC3, 1.0f / 255, 0);
processNet("dnn/yolov4-tiny.weights", "dnn/yolov4-tiny.cfg", "", inp);
}
PERF_TEST_P_(DNNTestNetwork, EAST_text_detection)
{
if (backend == DNN_BACKEND_HALIDE)

7
modules/dnn/src/layers/slice_layer.cpp
View File

@ -202,6 +202,10 @@ public:
#ifdef HAVE_OPENCL
bool forward_ocl(InputArrayOfArrays inputs_, OutputArrayOfArrays outputs_, OutputArrayOfArrays internals_)
{
#if 1
// TODO fix that (brokes YOLOv4-tiny)
return false;
#else
std::vector<UMat> inputs;
std::vector<UMat> outputs;
@ -244,7 +248,8 @@ public:
}
return true;
}
#endif
}
#endif
void forward(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr, OutputArrayOfArrays internals_arr) CV_OVERRIDE

8
modules/dnn/test/test_common.impl.hpp
View File

@ -96,9 +96,12 @@ void normAssertDetections(
const char *comment /*= ""*/, double confThreshold /*= 0.0*/,
double scores_diff /*= 1e-5*/, double boxes_iou_diff /*= 1e-4*/)
{
ASSERT_FALSE(testClassIds.empty()) << "No detections";
std::vector<bool> matchedRefBoxes(refBoxes.size(), false);
std::vector<double> refBoxesIoUDiff(refBoxes.size(), 1.0);
for (int i = 0; i < testBoxes.size(); ++i)
{
//cout << "Test[i=" << i << "]: score=" << testScores[i] << " id=" << testClassIds[i] << " box " << testBoxes[i] << endl;
double testScore = testScores[i];
if (testScore < confThreshold)
continue;
@ -115,6 +118,7 @@ void normAssertDetections(
double interArea = (testBox & refBoxes[j]).area();
double iou = interArea / (testBox.area() + refBoxes[j].area() - interArea);
topIoU = std::max(topIoU, iou);
refBoxesIoUDiff[j] = std::min(refBoxesIoUDiff[j], 1.0f - iou);
if (1.0 - iou < boxes_iou_diff)
{
matched = true;
@ -137,7 +141,9 @@ void normAssertDetections(
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;
refClassIds[i], refScores[i]) << refBoxes[i]
<< " IoU diff: " << refBoxesIoUDiff[i]
<< std::endl;
EXPECT_LE(refScores[i], confThreshold) << comment;
}
}

66
modules/dnn/test/test_darknet_importer.cpp
View File

@ -254,6 +254,13 @@ public:
}
if (cvIsNaN(iouDiff))
{
if (b == 0)
std::cout << "Skip accuracy checks" << std::endl;
continue;
}
normAssertDetections(refClassIds[b], refConfidences[b], refBoxes[b], nms_classIds,
nms_confidences, nms_boxes, format("batch size %d, sample %d\n", batch_size, b).c_str(), confThreshold, scoreDiff, iouDiff);
}
@ -449,7 +456,7 @@ TEST_P(Test_Darknet_nets_async, Accuracy)
}
INSTANTIATE_TEST_CASE_P(/**/, Test_Darknet_nets_async, Combine(
Values("yolo-voc", "tiny-yolo-voc", "yolov3", "yolov4"),
Values("yolo-voc", "tiny-yolo-voc", "yolov3", "yolov4", "yolov4-tiny"),
dnnBackendsAndTargets()
));
@ -587,6 +594,63 @@ TEST_P(Test_Darknet_nets, YOLOv4)
}
}
TEST_P(Test_Darknet_nets, YOLOv4_tiny)
{
applyTestTag(
target == DNN_TARGET_CPU ? CV_TEST_TAG_MEMORY_512MB : CV_TEST_TAG_MEMORY_1GB
);
const double confThreshold = 0.5;
// batchId, classId, confidence, left, top, right, bottom
const int N0 = 2;
const int N1 = 3;
static const float ref_[/* (N0 + N1) * 7 */] = {
0, 7, 0.85935f, 0.593484f, 0.141211f, 0.920356f, 0.291593f,
0, 16, 0.795188f, 0.169207f, 0.386886f, 0.423753f, 0.933004f,
1, 2, 0.996832f, 0.653802f, 0.464573f, 0.815193f, 0.653292f,
1, 2, 0.963325f, 0.451151f, 0.458915f, 0.496255f, 0.52241f,
1, 0, 0.926244f, 0.194851f, 0.361743f, 0.260277f, 0.632364f,
};
Mat ref(N0 + N1, 7, CV_32FC1, (void*)ref_);
double scoreDiff = 0.01f;
double iouDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.15 : 0.01f;
std::string config_file = "yolov4-tiny.cfg";
std::string weights_file = "yolov4-tiny.weights";
#if defined(INF_ENGINE_RELEASE)
if (target == DNN_TARGET_MYRIAD) // bad accuracy
iouDiff = std::numeric_limits<double>::quiet_NaN();
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_OPENCL)
iouDiff = std::numeric_limits<double>::quiet_NaN();
if ((backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 ||
backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH) && DNN_TARGET_OPENCL_FP16)
iouDiff = std::numeric_limits<double>::quiet_NaN();
#endif
{
SCOPED_TRACE("batch size 1");
testDarknetModel(config_file, weights_file, ref.rowRange(0, N0), scoreDiff, iouDiff, confThreshold);
}
{
SCOPED_TRACE("batch size 2");
testDarknetModel(config_file, weights_file, ref, scoreDiff, iouDiff, confThreshold);
}
#if defined(INF_ENGINE_RELEASE)
if (target == DNN_TARGET_MYRIAD) // bad accuracy
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_OPENCL)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
if ((backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 ||
backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH) && DNN_TARGET_OPENCL_FP16)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
#endif
}
INSTANTIATE_TEST_CASE_P(/**/, Test_Darknet_nets, dnnBackendsAndTargets());