Chiebot-Mirror
/
opencv
mirror of https://github.com/opencv/opencv.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

Merge pull request #16010 from YashasSamaga:cuda4dnn-fp16-tests

Browse Source 
* enable tests for DNN_TARGET_CUDA_FP16

* disable deconvolution tests

* disable shortcut tests

* fix typos and some minor changes

* dnn(test): skip CUDA FP16 test too (run_pool_max)
pull/16211/head
Yashas Samaga B L 5 years ago committed by Alexander Alekhin
parent 4de1efd3c8
commit 1fac1421e5
10 changed files with 461 additions and 107 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 126
      modules/dnn/test/test_backends.cpp
  2. 63
      modules/dnn/test/test_caffe_importer.cpp
  3. 5
      modules/dnn/test/test_common.impl.hpp
  4. 42
      modules/dnn/test/test_darknet_importer.cpp
  5. 27
      modules/dnn/test/test_halide_layers.cpp
  6. 43
      modules/dnn/test/test_layers.cpp
  7. 55
      modules/dnn/test/test_model.cpp
  8. 35
      modules/dnn/test/test_onnx_importer.cpp
  9. 94
      modules/dnn/test/test_tf_importer.cpp
  10. 78
      modules/dnn/test/test_torch_importer.cpp

126
modules/dnn/test/test_backends.cpp
Unescape View File

@ -168,6 +168,8 @@ TEST_P(DNNTestNetwork, ENet)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16)
applyTestTag(CV_TEST_TAG_DNN_SKIP_OPENCL_FP16);
if (backend == DNN_BACKEND_CUDA && target == DNN_TARGET_CUDA_FP16)
applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA_FP16);
processNet("dnn/Enet-model-best.net", "", Size(512, 512), "l367_Deconvolution",
target == DNN_TARGET_OPENCL ? "dnn/halide_scheduler_opencl_enet.yml" :
"dnn/halide_scheduler_enet.yml",
@ -182,11 +184,11 @@ TEST_P(DNNTestNetwork, MobileNet_SSD_Caffe)
applyTestTag(CV_TEST_TAG_DNN_SKIP_HALIDE);
Mat sample = imread(findDataFile("dnn/street.png"));
Mat inp = blobFromImage(sample, 1.0f / 127.5, Size(300, 300), Scalar(127.5, 127.5, 127.5), false);
float diffScores = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 1.5e-2 : 0.0;
float diffSquares = (target == DNN_TARGET_MYRIAD) ? 0.063 : 0.0;
float scoreDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 1.5e-2 : 0.0;
float iouDiff = (target == DNN_TARGET_MYRIAD) ? 0.063 : 0.0;
float detectionConfThresh = (target == DNN_TARGET_MYRIAD) ? 0.252 : FLT_MIN;
processNet("dnn/MobileNetSSD_deploy.caffemodel", "dnn/MobileNetSSD_deploy.prototxt",
inp, "detection_out", "", diffScores, diffSquares, detectionConfThresh);
inp, "detection_out", "", scoreDiff, iouDiff, detectionConfThresh);
expectNoFallbacksFromIE(net);
}
@ -201,10 +203,19 @@ TEST_P(DNNTestNetwork, MobileNet_SSD_Caffe_Different_Width_Height)
#endif
Mat sample = imread(findDataFile("dnn/street.png"));
Mat inp = blobFromImage(sample, 1.0f / 127.5, Size(300, 560), Scalar(127.5, 127.5, 127.5), false);
float diffScores = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.029 : 0.0;
float diffSquares = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.09 : 0.0;
float scoreDiff = 0.0, iouDiff = 0.0;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
scoreDiff = 0.029;
iouDiff = 0.09;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
scoreDiff = 0.03;
iouDiff = 0.08;
}
processNet("dnn/MobileNetSSD_deploy.caffemodel", "dnn/MobileNetSSD_deploy.prototxt",
inp, "detection_out", "", diffScores, diffSquares);
inp, "detection_out", "", scoreDiff, iouDiff);
expectNoFallbacksFromIE(net);
}
@ -216,11 +227,20 @@ TEST_P(DNNTestNetwork, MobileNet_SSD_v1_TensorFlow)
Mat sample = imread(findDataFile("dnn/street.png"));
Mat inp = blobFromImage(sample, 1.0f, Size(300, 300), Scalar(), false);
float l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.095 : 0.0;
float lInf = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.09 : 0.0;
float detectionConfThresh = (target == DNN_TARGET_MYRIAD) ? 0.216 : 0.2;
float scoreDiff = 0.0, iouDiff = 0.0;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
scoreDiff = 0.095;
iouDiff = 0.09;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
scoreDiff = 0.007;
iouDiff = 0.08;
}
processNet("dnn/ssd_mobilenet_v1_coco_2017_11_17.pb", "dnn/ssd_mobilenet_v1_coco_2017_11_17.pbtxt",
inp, "detection_out", "", l1, lInf, detectionConfThresh);
inp, "detection_out", "", scoreDiff, iouDiff, detectionConfThresh);
expectNoFallbacksFromIE(net);
}
@ -240,10 +260,19 @@ TEST_P(DNNTestNetwork, MobileNet_SSD_v1_TensorFlow_Different_Width_Height)
Mat sample = imread(findDataFile("dnn/street.png"));
Mat inp = blobFromImage(sample, 1.0f, Size(300, 560), Scalar(), false);
float l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.012 : 0.0;
float lInf = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.06 : 0.0;
float scoreDiff = 0.0, iouDiff = 0.0;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
scoreDiff = 0.012;
iouDiff = 0.06;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
scoreDiff = 0.007;
iouDiff = 0.06;
}
processNet("dnn/ssd_mobilenet_v1_coco_2017_11_17.pb", "dnn/ssd_mobilenet_v1_coco_2017_11_17.pbtxt",
inp, "detection_out", "", l1, lInf);
inp, "detection_out", "", scoreDiff, iouDiff);
expectNoFallbacksFromIE(net);
}
@ -255,10 +284,19 @@ TEST_P(DNNTestNetwork, MobileNet_SSD_v2_TensorFlow)
Mat sample = imread(findDataFile("dnn/street.png"));
Mat inp = blobFromImage(sample, 1.0f, Size(300, 300), Scalar(), false);
float l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.013 : 2e-5;
float lInf = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.062 : 0.0;
float scoreDiff = 2e-5, iouDiff = 0.0;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
scoreDiff = 0.013;
iouDiff = 0.062;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
scoreDiff = 0.02;
iouDiff = 0.07;
}
processNet("dnn/ssd_mobilenet_v2_coco_2018_03_29.pb", "dnn/ssd_mobilenet_v2_coco_2018_03_29.pbtxt",
inp, "detection_out", "", l1, lInf, 0.25);
inp, "detection_out", "", scoreDiff, iouDiff, 0.25);
expectNoFallbacksFromIE(net);
}
@ -268,12 +306,25 @@ TEST_P(DNNTestNetwork, SSD_VGG16)
CV_TEST_TAG_DEBUG_VERYLONG);
if (backend == DNN_BACKEND_HALIDE && target == DNN_TARGET_CPU)
applyTestTag(CV_TEST_TAG_DNN_SKIP_HALIDE); // TODO HALIDE_CPU
double scoreThreshold = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.0325 : 0.0;
const float lInf = (target == DNN_TARGET_MYRIAD) ? 0.032 : 0.0;
Mat sample = imread(findDataFile("dnn/street.png"));
Mat inp = blobFromImage(sample, 1.0f, Size(300, 300), Scalar(), false);
float scoreDiff = 0.0, iouDiff = 0.0;
if (target == DNN_TARGET_OPENCL_FP16)
{
scoreDiff = 0.0325;
}
else if (target == DNN_TARGET_MYRIAD)
{
scoreDiff = 0.0325;
iouDiff = 0.032;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
scoreDiff = 0.03;
}
processNet("dnn/VGG_ILSVRC2016_SSD_300x300_iter_440000.caffemodel",
"dnn/ssd_vgg16.prototxt", inp, "detection_out", "", scoreThreshold, lInf);
"dnn/ssd_vgg16.prototxt", inp, "detection_out", "", scoreDiff, iouDiff);
expectNoFallbacksFromIE(net);
}
@ -384,10 +435,19 @@ TEST_P(DNNTestNetwork, Inception_v2_SSD_TensorFlow)
applyTestTag(CV_TEST_TAG_DNN_SKIP_HALIDE);
Mat sample = imread(findDataFile("dnn/street.png"));
Mat inp = blobFromImage(sample, 1.0f, Size(300, 300), Scalar(), false);
float l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.015 : 0.0;
float lInf = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.0731 : 0.0;
float scoreDiff = 0.0, iouDiff = 0.0;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
scoreDiff = 0.015;
iouDiff = 0.0731;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
scoreDiff = 0.015;
iouDiff = 0.08;
}
processNet("dnn/ssd_inception_v2_coco_2017_11_17.pb", "dnn/ssd_inception_v2_coco_2017_11_17.pbtxt",
inp, "detection_out", "", l1, lInf);
inp, "detection_out", "", scoreDiff, iouDiff);
expectNoFallbacksFromIE(net);
}
@ -400,11 +460,18 @@ TEST_P(DNNTestNetwork, DenseNet_121)
float l1 = 0.0, lInf = 0.0;
if (target == DNN_TARGET_OPENCL_FP16)
{
l1 = 2e-2; lInf = 9e-2;
l1 = 2e-2;
lInf = 9e-2;
}
else if (target == DNN_TARGET_MYRIAD)
{
l1 = 0.1; lInf = 0.6;
l1 = 0.1;
lInf = 0.6;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
l1 = 0.008;
lInf = 0.05;
}
processNet("dnn/DenseNet_121.caffemodel", "dnn/DenseNet_121.prototxt", Size(224, 224), "", "", l1, lInf);
if (target != DNN_TARGET_MYRIAD || getInferenceEngineVPUType() != CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X)
@ -431,8 +498,17 @@ TEST_P(DNNTestNetwork, FastNeuralStyle_eccv16)
Mat img = imread(findDataFile("dnn/googlenet_1.png"));
Mat inp = blobFromImage(img, 1.0, Size(320, 240), Scalar(103.939, 116.779, 123.68), false, false);
// Output image has values in range [-143.526, 148.539].
float l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.4 : 4e-5;
float lInf = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 7.45 : 2e-3;
float l1 = 4e-5, lInf = 2e-3;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
l1 = 0.4;
lInf = 7.45;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
l1 = 0.3;
lInf = 7.2;
}
processNet("dnn/fast_neural_style_eccv16_starry_night.t7", "", inp, "", "", l1, lInf);
#if defined(HAVE_INF_ENGINE) && INF_ENGINE_VER_MAJOR_GE(2019010000)
expectNoFallbacksFromIE(net);

63
modules/dnn/test/test_caffe_importer.cpp
Unescape View File

@ -150,8 +150,17 @@ TEST_P(Test_Caffe_nets, Axpy)
}
}
}
float l1 = (target == DNN_TARGET_OPENCL_FP16) ? 2e-4 : 1e-5;
float lInf = (target == DNN_TARGET_OPENCL_FP16) ? 1e-3 : 1e-4;
float l1 = 1e-5, lInf = 1e-4;
if (target == DNN_TARGET_OPENCL_FP16)
{
l1 = 2e-4;
lInf = 1e-3;
}
else if(target == DNN_TARGET_CUDA_FP16)
{
l1 = 0.0002;
lInf = 0.0007;
}
normAssert(ref, out, "", l1, lInf);
}
@ -287,8 +296,17 @@ TEST_P(Reproducibility_MobileNet_SSD, Accuracy)
ASSERT_EQ(out.size[2], 100);
const float scores_diff = (targetId == DNN_TARGET_OPENCL_FP16 || targetId == DNN_TARGET_MYRIAD) ? 1.5e-2 : 1e-5;
const float boxes_iou_diff = (targetId == DNN_TARGET_OPENCL_FP16 || targetId == DNN_TARGET_MYRIAD) ? 6.3e-2 : 1e-4;
float scores_diff = 1e-5, boxes_iou_diff = 1e-4;
if (targetId == DNN_TARGET_OPENCL_FP16 || targetId == DNN_TARGET_MYRIAD)
{
scores_diff = 1.5e-2;
boxes_iou_diff = 6.3e-2;
}
else if (targetId == DNN_TARGET_CUDA_FP16)
{
scores_diff = 0.015;
boxes_iou_diff = 0.07;
}
Mat ref = blobFromNPY(_tf("mobilenet_ssd_caffe_out.npy"));
normAssertDetections(ref, out, "", FLT_MIN, scores_diff, boxes_iou_diff);
@ -477,11 +495,21 @@ TEST_P(Test_Caffe_nets, Colorization)
Mat out = net.forward();
// Reference output values are in range [-29.1, 69.5]
double l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.25 : 4e-4;
double lInf = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 5.3 : 3e-3;
if (target == DNN_TARGET_MYRIAD && getInferenceEngineVPUType() == CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X)
double l1 = 4e-4, lInf = 3e-3;
if (target == DNN_TARGET_OPENCL_FP16)
{
l1 = 0.5; lInf = 11;
l1 = 0.25;
lInf = 5.3;
}
else if (target == DNN_TARGET_MYRIAD)
{
l1 = (getInferenceEngineVPUType() == CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X) ? 0.5 : 0.25;
lInf = (getInferenceEngineVPUType() == CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X) ? 11 : 5.3;
}
else if(target == DNN_TARGET_CUDA_FP16)
{
l1 = 0.21;
lInf = 4.5;
}
normAssert(out, ref, "", l1, lInf);
expectNoFallbacksFromIE(net);
@ -518,6 +546,10 @@ TEST_P(Test_Caffe_nets, DenseNet_121)
{
l1 = 0.11; lInf = 0.5;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
l1 = 0.04; lInf = 0.2;
}
normAssert(outs[0], ref, "", l1, lInf);
if (target != DNN_TARGET_MYRIAD || getInferenceEngineVPUType() != CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X)
expectNoFallbacksFromIE(model);
@ -663,6 +695,8 @@ TEST_P(Test_Caffe_nets, FasterRCNN_zf)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16);
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_MYRIAD)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD);
if (target == DNN_TARGET_CUDA_FP16)
applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA_FP16);
static Mat ref = (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);
@ -680,8 +714,17 @@ TEST_P(Test_Caffe_nets, RFCN)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16);
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_MYRIAD)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD);
double scoreDiff = (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16) ? 4e-3 : default_l1;
double iouDiff = (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16) ? 8e-2 : default_lInf;
float scoreDiff = default_l1, iouDiff = default_lInf;
if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16)
{
scoreDiff = 4e-3;
iouDiff = 8e-2;
}
if (target == DNN_TARGET_CUDA_FP16)
{
scoreDiff = 0.0034;
iouDiff = 0.11;
}
static Mat ref = (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);
testFaster("rfcn_pascal_voc_resnet50.prototxt", "resnet50_rfcn_final.caffemodel", ref, scoreDiff, iouDiff);

5
modules/dnn/test/test_common.impl.hpp
Unescape View File

@ -239,9 +239,8 @@ testing::internal::ParamGenerator< tuple<Backend, Target> > dnnBackendsAndTarget
#ifdef HAVE_CUDA
if(withCUDA)
{
//for (auto target : getAvailableTargets(DNN_BACKEND_CUDA))
// targets.push_back(make_tuple(DNN_BACKEND_CUDA, target));
targets.push_back(make_tuple(DNN_BACKEND_CUDA, DNN_TARGET_CUDA));
for (auto target : getAvailableTargets(DNN_BACKEND_CUDA))
targets.push_back(make_tuple(DNN_BACKEND_CUDA, target));
}
#endif

42
modules/dnn/test/test_darknet_importer.cpp
Unescape View File

@ -320,9 +320,18 @@ TEST_P(Test_Darknet_nets, YoloVoc)
1, 6, 0.667770f, 0.446555f, 0.453578f, 0.499986f, 0.519167f, // a car
1, 6, 0.844947f, 0.637058f, 0.460398f, 0.828508f, 0.66427f); // a car
double scoreDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 1e-2 : 8e-5;
double iouDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.018 : 3e-4;
double nmsThreshold = (target == DNN_TARGET_MYRIAD) ? 0.397 : 0.4;
double scoreDiff = 8e-5, iouDiff = 3e-4;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
scoreDiff = 1e-2;
iouDiff = 0.018;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
scoreDiff = 0.03;
iouDiff = 0.018;
}
std::string config_file = "yolo-voc.cfg";
std::string weights_file = "yolo-voc.weights";
@ -353,8 +362,17 @@ TEST_P(Test_Darknet_nets, TinyYoloVoc)
1, 6, 0.651450f, 0.460526f, 0.458019f, 0.522527f, 0.5341f, // a car
1, 6, 0.928758f, 0.651024f, 0.463539f, 0.823784f, 0.654998f); // a car
double scoreDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 8e-3 : 8e-5;
double iouDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.018 : 3e-4;
double scoreDiff = 8e-5, iouDiff = 3e-4;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
scoreDiff = 8e-3;
iouDiff = 0.018;
}
else if(target == DNN_TARGET_CUDA_FP16)
{
scoreDiff = 0.008;
iouDiff = 0.02;
}
std::string config_file = "tiny-yolo-voc.cfg";
std::string weights_file = "tiny-yolo-voc.weights";
@ -453,9 +471,17 @@ TEST_P(Test_Darknet_nets, YOLOv3)
1, 2, 0.989633f, 0.450719f, 0.463353f, 0.496305f, 0.522258f, // a car
1, 2, 0.997412f, 0.647584f, 0.459939f, 0.821038f, 0.663947f); // a car
double scoreDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.006 : 8e-5;
double iouDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.018 : 3e-4;
double scoreDiff = 8e-5, iouDiff = 3e-4;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
scoreDiff = 0.006;
iouDiff = 0.018;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
scoreDiff = 0.04;
iouDiff = 0.03;
}
std::string config_file = "yolov3.cfg";
std::string weights_file = "yolov3.weights";
@ -501,6 +527,8 @@ INSTANTIATE_TEST_CASE_P(/**/, Test_Darknet_nets, dnnBackendsAndTargets());
TEST_P(Test_Darknet_layers, shortcut)
{
if (backend == DNN_BACKEND_CUDA)
applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA);
testDarknetLayer("shortcut");
testDarknetLayer("shortcut_leaky");
testDarknetLayer("shortcut_unequal");

27
modules/dnn/test/test_halide_layers.cpp
Unescape View File

@ -16,7 +16,7 @@ using namespace cv;
using namespace cv::dnn;
using namespace testing;
static void test(Mat& input, Net& net, Backend backendId, Target targetId, bool skipCheck = false, bool randInput = true)
static void test(Mat& input, Net& net, Backend backendId, Target targetId, bool skipCheck = false, bool randInput = true, double l1 = 0.0, double lInf = 0.0)
{
DNNTestLayer::checkBackend(backendId, targetId);
if (randInput)
@ -33,8 +33,12 @@ static void test(Mat& input, Net& net, Backend backendId, Target targetId, bool
if (skipCheck)
return;
double l1, lInf;
DNNTestLayer::getDefaultThresholds(backendId, targetId, &l1, &lInf);
double default_l1, default_lInf;
DNNTestLayer::getDefaultThresholds(backendId, targetId, &default_l1, &default_lInf);
if (l1 == 0.0)
l1 = default_l1;
if (lInf == 0.0)
lInf = default_lInf;
#if 0
std::cout << "l1=" << l1 << " lInf=" << lInf << std::endl;
std::cout << outputDefault.reshape(1, outputDefault.total()).t() << std::endl;
@ -43,11 +47,11 @@ static void test(Mat& input, Net& net, Backend backendId, Target targetId, bool
normAssert(outputDefault, outputHalide, "", l1, lInf);
}
static void test(LayerParams& params, Mat& input, Backend backendId, Target targetId, bool skipCheck = false)
static void test(LayerParams& params, Mat& input, Backend backendId, Target targetId, bool skipCheck = false, double l1 = 0.0, double lInf = 0.0)
{
Net net;
net.addLayerToPrev(params.name, params.type, params);
test(input, net, backendId, targetId, skipCheck);
test(input, net, backendId, targetId, skipCheck, true, l1, lInf);
}
static inline testing::internal::ParamGenerator<tuple<Backend, Target> > dnnBackendsAndTargetsWithHalide()
@ -174,6 +178,9 @@ TEST_P(Deconvolution, Accuracy)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD_X);
#endif
if (targetId == DNN_TARGET_CUDA_FP16)
applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA_FP16);
int sz[] = {inChannels, outChannels / group, kernel.height, kernel.width};
Mat weights(4, &sz[0], CV_32F);
randu(weights, -1.0f, 1.0f);
@ -414,7 +421,11 @@ TEST_P(FullyConnected, Accuracy)
int sz[] = {1, inChannels, inSize.height, inSize.width};
Mat input(4, &sz[0], CV_32F);
test(lp, input, backendId, targetId);
double l1 = 0.0;
if (targetId == DNN_TARGET_CUDA_FP16)
l1 = 0.015;
test(lp, input, backendId, targetId, false, true, l1);
}
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, FullyConnected, Combine(
@ -497,7 +508,7 @@ TEST_P(Test_Halide_layers, MaxPoolUnpool)
////////////////////////////////////////////////////////////////////////////////
static const int kNumChannels = 3;
void testInPlaceActivation(LayerParams& lp, Backend backendId, Target targetId)
void testInPlaceActivation(LayerParams& lp, Backend backendId, Target targetId, double l1 = 0.0, double lInf = 0.0)
{
EXPECT_FALSE(lp.name.empty());
@ -517,7 +528,7 @@ void testInPlaceActivation(LayerParams& lp, Backend backendId, Target targetId)
int sz[] = {1, kNumChannels, 10, 10};
Mat input(4, &sz[0], CV_32F);
test(input, net, backendId, targetId);
test(input, net, backendId, targetId, false, true, l1, lInf);
}
typedef TestWithParam<tuple<bool, bool, float, tuple<Backend, Target> > > BatchNorm;

43
modules/dnn/test/test_layers.cpp
Unescape View File

@ -141,6 +141,8 @@ TEST_P(Test_Caffe_layers, Convolution)
TEST_P(Test_Caffe_layers, DeConvolution)
{
if(target == DNN_TARGET_CUDA_FP16)
applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA_FP16);
testLayerUsingCaffeModels("layer_deconvolution", true, false);
}
@ -372,7 +374,13 @@ TEST_P(Test_Caffe_layers, Conv_Elu)
net.setPreferableTarget(target);
Mat out = net.forward();
normAssert(ref, out, "", default_l1, default_lInf);
double l1 = default_l1, lInf = default_lInf;
if (target == DNN_TARGET_CUDA_FP16)
{
l1 = 0.0002;
lInf = 0.0005;
}
normAssert(ref, out, "", l1, lInf);
}
class Layer_LSTM_Test : public ::testing::Test
@ -843,6 +851,11 @@ TEST_P(Test_Caffe_layers, PriorBox_repeated)
double l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 1e-3 : 1e-5;
double lInf = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 1e-3 : 1e-4;
if (target == DNN_TARGET_CUDA_FP16)
{
l1 = 7e-5;
lInf = 0.0005;
}
normAssert(out, ref, "", l1, lInf);
}
@ -876,7 +889,9 @@ TEST_P(Test_Caffe_layers, PriorBox_squares)
0.25, 0.0, 1.0, 1.0,
0.1f, 0.1f, 0.2f, 0.2f,
0.1f, 0.1f, 0.2f, 0.2f);
double l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 2e-5 : 1e-5;
double l1 = 1e-5;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD || target == DNN_TARGET_CUDA_FP16)
l1 = 2e-5;
normAssert(out.reshape(1, 4), ref, "", l1);
}
@ -1225,6 +1240,11 @@ TEST_P(Test_DLDT_two_inputs, as_backend)
// Output values are in range [0, 637.5].
double l1 = (targetId == DNN_TARGET_OPENCL_FP16 || targetId == DNN_TARGET_MYRIAD) ? 0.06 : 1e-6;
double lInf = (targetId == DNN_TARGET_OPENCL_FP16 || targetId == DNN_TARGET_MYRIAD) ? 0.3 : 1e-5;
if (targetId == DNN_TARGET_CUDA_FP16)
{
l1 = 0.06;
lInf = 0.3;
}
normAssert(out, ref, "", l1, lInf);
}
@ -1537,8 +1557,17 @@ TEST_P(Layer_Test_ShuffleChannel, Accuracy)
net.setPreferableTarget(targetId);
Mat out = net.forward();
double l1 = (targetId == DNN_TARGET_OPENCL_FP16) ? 5e-2 : 1e-5;
double lInf = (targetId == DNN_TARGET_OPENCL_FP16) ? 7e-2 : 1e-4;
double l1 = 1e-5, lInf = 1e-4;
if (targetId == DNN_TARGET_OPENCL_FP16)
{
l1 = 5e-2;
lInf = 7e-2;
}
else if (targetId == DNN_TARGET_CUDA_FP16)
{
l1 = 0.06;
lInf = 0.07;
}
for (int n = 0; n < inpShapeVec[0]; ++n)
{
for (int c = 0; c < inpShapeVec[1]; ++c)
@ -1593,6 +1622,9 @@ TEST_P(Layer_Test_Eltwise_unequal, accuracy_input_0_truncate)
int backendId = get<0>(get<1>(GetParam()));
int targetId = get<1>(get<1>(GetParam()));
if (backendId == DNN_BACKEND_CUDA)
applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA);
Net net;
LayerParams lp;
lp.type = "Eltwise";
@ -1656,6 +1688,9 @@ TEST_P(Layer_Test_Eltwise_unequal, accuracy_input_0)
int backendId = get<0>(get<1>(GetParam()));
int targetId = get<1>(get<1>(GetParam()));
if (backendId == DNN_BACKEND_CUDA)
applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA);
Net net;
LayerParams lp;
lp.type = "Eltwise";

55
modules/dnn/test/test_model.cpp
Unescape View File

@ -157,9 +157,13 @@ TEST_P(Test_Model, DetectRegion)
bool swapRB = true;
double confThreshold = 0.24;
double scoreDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 1e-2 : 8e-5;
double iouDiff = (target == DNN_TARGET_MYRIAD || target == DNN_TARGET_OPENCL_FP16) ? 1.6e-2 : 1e-5;
double nmsThreshold = (target == DNN_TARGET_MYRIAD) ? 0.397 : 0.4;
double scoreDiff = 8e-5, iouDiff = 1e-5;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD || target == DNN_TARGET_CUDA_FP16)
{
scoreDiff = 1e-2;
iouDiff = 1.6e-2;
}
testDetectModel(weights_file, config_file, img_path, refClassIds, refConfidences,
refBoxes, scoreDiff, iouDiff, confThreshold, nmsThreshold, size,
@ -188,11 +192,15 @@ TEST_P(Test_Model, DetectionOutput)
Scalar mean = Scalar(102.9801, 115.9465, 122.7717);
Size size{800, 600};
double scoreDiff = (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16) ?
4e-3 : default_l1;
double iouDiff = (target == DNN_TARGET_OPENCL_FP16) ? 1.8e-1 : 1e-5;
double scoreDiff = default_l1, iouDiff = 1e-5;
float confThreshold = 0.8;
double nmsThreshold = 0.0;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_CUDA_FP16)
{
if (backend == DNN_BACKEND_OPENCV)
scoreDiff = 4e-3;
iouDiff = 1.8e-1;
}
testDetectModel(weights_file, config_file, img_path, refClassIds, refConfidences, refBoxes,
scoreDiff, iouDiff, confThreshold, nmsThreshold, size, mean);
@ -232,10 +240,22 @@ TEST_P(Test_Model, DetectionMobilenetSSD)
double scale = 1.0 / 127.5;
Size size{300, 300};
double scoreDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 1.7e-2 : 1e-5;
double iouDiff = (target == DNN_TARGET_OPENCL_FP16 || (target == DNN_TARGET_MYRIAD &&
getInferenceEngineVPUType() == CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X)) ? 6.91e-2 : 1e-5;
double scoreDiff = 1e-5, iouDiff = 1e-5;
if (target == DNN_TARGET_OPENCL_FP16)
{
scoreDiff = 1.7e-2;
iouDiff = 6.91e-2;
}
else if (target == DNN_TARGET_MYRIAD)
{
scoreDiff = 1.7e-2;
if (getInferenceEngineVPUType() == CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X)
iouDiff = 6.91e-2;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
scoreDiff = 4e-4;
}
float confThreshold = FLT_MIN;
double nmsThreshold = 0.0;
@ -263,6 +283,10 @@ TEST_P(Test_Model, Keypoints_pose)
Scalar mean = Scalar(128, 128, 128);
bool swapRB = false;
// Ref. Range: [58.6875, 508.625]
if (target == DNN_TARGET_CUDA_FP16)
norm = 20; // l1 = 1.5, lInf = 20
testKeypointsModel(weights, "", inp, exp, norm, size, mean, scale, swapRB);
}
@ -283,8 +307,11 @@ TEST_P(Test_Model, Keypoints_face)
Scalar mean = Scalar();
bool swapRB = false;
testKeypointsModel(weights, "", inp, exp, norm, size, mean, scale, swapRB);
// Ref. Range: [-1.1784188, 1.7758257]
if (target == DNN_TARGET_CUDA_FP16)
norm = 0.004; // l1 = 0.0006, lInf = 0.004
testKeypointsModel(weights, "", inp, exp, norm, size, mean, scale, swapRB);
}
TEST_P(Test_Model, Detection_normalized)
@ -301,10 +328,14 @@ TEST_P(Test_Model, Detection_normalized)
double scale = 1.0 / 127.5;
Size size{300, 300};
double scoreDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 5e-3 : 1e-5;
double iouDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.09 : 1e-5;
double scoreDiff = 1e-5, iouDiff = 1e-5;
float confThreshold = FLT_MIN;
double nmsThreshold = 0.0;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD || target == DNN_TARGET_CUDA_FP16)
{
scoreDiff = 5e-3;
iouDiff = 0.09;
}
testDetectModel(weights_file, config_file, img_path, refClassIds, refConfidences, refBoxes,
scoreDiff, iouDiff, confThreshold, nmsThreshold, size, mean, scale);
}

35
modules/dnn/test/test_onnx_importer.cpp
Unescape View File

@ -590,8 +590,17 @@ TEST_P(Test_ONNX_nets, TinyYolov2)
#endif
// output range: [-11; 8]
double l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.017 : default_l1;
double lInf = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.14 : default_lInf;
double l1 = default_l1, lInf = default_lInf;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
l1 = 0.017;
lInf = 0.14;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
l1 = 0.018;
lInf = 0.16;
}
testONNXModels("tiny_yolo2", pb, l1, lInf);
}
@ -620,17 +629,23 @@ TEST_P(Test_ONNX_nets, LResNet100E_IR)
if (target == DNN_TARGET_MYRIAD) applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
}
double l1 = default_l1;
double lInf = default_lInf;
double l1 = default_l1, lInf = default_lInf;
// output range: [-3; 3]
if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16) {
if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16)
{
l1 = 0.009;
lInf = 0.035;
}
else if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_CPU) {
else if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_CPU)
{
l1 = 4.6e-5;
lInf = 1.9e-4;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
l1 = 0.008;
lInf = 0.04;
}
testONNXModels("LResNet100E_IR", pb, l1, lInf);
}
@ -747,8 +762,12 @@ TEST_P(Test_ONNX_nets, Resnet34_kinetics)
net.setPreferableTarget(target);
// output range [-5, 11]
float l1 = 0.0013;
float lInf = 0.009;
float l1 = 0.0013, lInf = 0.009;
if (target == DNN_TARGET_CUDA_FP16)
{
l1 = 0.008;
lInf = 0.04;
}
checkBackend(&input0, &ref0);
net.setInput(input0);

94
modules/dnn/test/test_tf_importer.cpp
Unescape View File

@ -225,8 +225,17 @@ TEST_P(Test_TensorFlow_layers, slim_batch_norm)
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
// Output values range: [-40.0597, 207.827]
double l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.041 : default_l1;
double lInf = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.33 : default_lInf;
double l1 = default_l1, lInf = default_lInf;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
l1 = 0.041;
lInf = 0.33;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
l1 = 0.005;
lInf = 0.33;
}
runTensorFlowNet("slim_batch_norm", false, l1, lInf);
}
@ -300,9 +309,8 @@ TEST_P(Test_TensorFlow_layers, AvePooling3D)
TEST_P(Test_TensorFlow_layers, deconvolution)
{
if(backend == DNN_BACKEND_CUDA)
applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA); /* bugged */
if (backend == DNN_BACKEND_CUDA)
applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA);
runTensorFlowNet("deconvolution");
runTensorFlowNet("deconvolution_same");
runTensorFlowNet("deconvolution_stride_2_same");
@ -428,8 +436,16 @@ TEST_P(Test_TensorFlow_nets, MobileNet_SSD)
net.setInput(inp);
Mat out = net.forward();
double scoreDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.0043 : default_l1;
double iouDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.037 : default_lInf;
double scoreDiff = default_l1, iouDiff = default_lInf;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
scoreDiff = 0.0043;
iouDiff = 0.037;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
iouDiff = 0.04;
}
normAssertDetections(ref, out, "", 0.2, scoreDiff, iouDiff);
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_RELEASE >= 2019010000
expectNoFallbacksFromIE(net);
@ -466,8 +482,17 @@ TEST_P(Test_TensorFlow_nets, Inception_v2_SSD)
0, 10, 0.95932811, 0.38349164, 0.32528657, 0.40387636, 0.39165527,
0, 10, 0.93973452, 0.66561931, 0.37841269, 0.68074018, 0.42907384);
double scoreDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.0097 : default_l1;
double iouDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.09 : default_lInf;
double scoreDiff = default_l1, iouDiff = default_lInf;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
scoreDiff = 0.0097;
iouDiff = 0.09;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
scoreDiff = 6e-3;
iouDiff = 0.05;
}
normAssertDetections(ref, out, "", 0.5, scoreDiff, iouDiff);
expectNoFallbacksFromIE(net);
}
@ -489,10 +514,18 @@ TEST_P(Test_TensorFlow_nets, MobileNet_v1_SSD)
Mat out = net.forward();
Mat ref = blobFromNPY(findDataFile("dnn/tensorflow/ssd_mobilenet_v1_coco_2017_11_17.detection_out.npy"));
float scoreDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.011 : 1.5e-5;
float iouDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.012 : 1e-3;
float scoreDiff = 1.5e-5, iouDiff = 1e-3;
float detectionConfThresh = (target == DNN_TARGET_MYRIAD) ? 0.35 : 0.3;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
scoreDiff = 0.011;
iouDiff = 0.012;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
scoreDiff = 0.006;
iouDiff = 0.01;
}
#if defined(INF_ENGINE_RELEASE)
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_MYRIAD &&
getInferenceEngineVPUType() == CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X)
@ -530,6 +563,9 @@ TEST_P(Test_TensorFlow_nets, Faster_RCNN)
if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16)
applyTestTag(CV_TEST_TAG_DNN_SKIP_OPENCL_FP16);
if (backend == DNN_BACKEND_CUDA && target == DNN_TARGET_CUDA_FP16)
applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA_FP16);
checkBackend();
double scoresDiff = backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 ? 2.9e-5 : 1e-5;
@ -574,8 +610,17 @@ TEST_P(Test_TensorFlow_nets, MobileNet_v1_SSD_PPN)
net.setInput(blob);
Mat out = net.forward();
double scoreDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.048 : 1.1e-5;
double iouDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.058 : default_lInf;
double scoreDiff = 1.1e-5, iouDiff = default_lInf;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
scoreDiff = 0.048;
iouDiff = 0.058;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
scoreDiff = 0.006;
iouDiff = 0.05;
}
normAssertDetections(ref, out, "", 0.45, scoreDiff, iouDiff);
expectNoFallbacksFromIE(net);
}
@ -604,8 +649,17 @@ TEST_P(Test_TensorFlow_nets, opencv_face_detector_uint8)
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);
double scoreDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 4e-3 : 3.4e-3;
double iouDiff = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.024 : 1e-2;
double scoreDiff = 3.4e-3, iouDiff = 1e-2;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
scoreDiff = 4e-3;
iouDiff = 0.024;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
scoreDiff = 4e-3;
iouDiff = 0.02;
}
normAssertDetections(ref, out, "", 0.9, scoreDiff, iouDiff);
expectNoFallbacksFromIE(net);
}
@ -673,6 +727,11 @@ TEST_P(Test_TensorFlow_nets, EAST_text_detection)
lInf_scores = 0.41;
l1_geometry = 0.28; lInf_geometry = 5.94;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
lInf_scores = 0.1;
l1_geometry = 0.3; lInf_geometry = 7;
}
else
{
l1_geometry = 1e-4, lInf_geometry = 3e-3;
@ -695,7 +754,8 @@ TEST_P(Test_TensorFlow_layers, fp16_weights)
runTensorFlowNet("fp16_padding_valid", false, l1, lInf);
// Reference output values are in range [0.0889, 1.651]
runTensorFlowNet("fp16_max_pool_even", false, (target == DNN_TARGET_MYRIAD) ? 0.003 : l1, lInf);
if (target == DNN_TARGET_MYRIAD) {
if (target == DNN_TARGET_MYRIAD)
{
l1 = 0.0041;
lInf = 0.024;
}

78
modules/dnn/test/test_torch_importer.cpp
Unescape View File

@ -112,8 +112,17 @@ public:
TEST_P(Test_Torch_layers, run_convolution)
{
// Output reference values are in range [23.4018, 72.0181]
double l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.08 : default_l1;
double lInf = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.42 : default_lInf;
double l1 = default_l1, lInf = default_lInf;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
l1 = 0.08;
lInf = 0.42;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
l1 = 0.08;
lInf = 0.5;
}
runTorchNet("net_conv", "", false, true, true, l1, lInf);
}
@ -121,7 +130,10 @@ TEST_P(Test_Torch_layers, run_pool_max)
{
if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16)
applyTestTag(CV_TEST_TAG_DNN_SKIP_OPENCL_FP16);
runTorchNet("net_pool_max", "", true);
if (target == DNN_TARGET_CUDA_FP16)
applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA_FP16);
double l1 = 0.0, lInf = 0.0;
runTorchNet("net_pool_max", "", true, false, true, l1, lInf);
}
TEST_P(Test_Torch_layers, run_pool_ave)
@ -145,9 +157,17 @@ TEST_P(Test_Torch_layers, run_reshape)
TEST_P(Test_Torch_layers, run_reshape_single_sample)
{
// Reference output values in range [14.4586, 18.4492].
runTorchNet("net_reshape_single_sample", "", false, false, true,
(target == DNN_TARGET_MYRIAD || target == DNN_TARGET_OPENCL_FP16) ? 0.033 : default_l1,
(target == DNN_TARGET_MYRIAD || target == DNN_TARGET_OPENCL_FP16) ? 0.05 : default_lInf);
double l1 = default_l1, lInf = default_lInf;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
l1 = 0.033;
lInf = 0.05;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
l1 = 0.01;
}
runTorchNet("net_reshape_single_sample", "", false, false, true, l1, lInf);
}
TEST_P(Test_Torch_layers, run_linear)
@ -164,8 +184,16 @@ TEST_P(Test_Torch_layers, run_concat)
TEST_P(Test_Torch_layers, run_depth_concat)
{
runTorchNet("net_depth_concat", "", false, true, true, 0.0,
target == DNN_TARGET_OPENCL_FP16 ? 0.021 : 0.0);
double lInf = 0.0;
if (target == DNN_TARGET_OPENCL_FP16)
{
lInf = 0.021;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
lInf = 0.03;
}
runTorchNet("net_depth_concat", "", false, true, true, 0.0, lInf);
}
TEST_P(Test_Torch_layers, run_deconv)
@ -211,9 +239,18 @@ TEST_P(Test_Torch_layers, net_conv_gemm_lrn)
{
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_MYRIAD)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
runTorchNet("net_conv_gemm_lrn", "", false, true, true,
target == DNN_TARGET_OPENCL_FP16 ? 0.046 : 0.0,
target == DNN_TARGET_OPENCL_FP16 ? 0.023 : 0.0);
double l1 = 0.0, lInf = 0.0;
if (target == DNN_TARGET_OPENCL_FP16)
{
l1 = 0.046;
lInf = 0.023;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
l1 = 0.0042;
lInf = 0.021;
}
runTorchNet("net_conv_gemm_lrn", "", false, true, true, l1, lInf);
}
TEST_P(Test_Torch_layers, net_inception_block)
@ -291,8 +328,17 @@ TEST_P(Test_Torch_nets, OpenFace_accuracy)
// Reference output values are in range [-0.17212, 0.263492]
// on Myriad problem layer: l4_Pooling - does not use pads_begin
float l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 2e-3 : 1e-5;
float lInf = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 5e-3 : 1e-3;
float l1 = 1e-5, lInf = 1e-3;
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
l1 = 2e-3;
lInf = 5e-3;
}
else if (target == DNN_TARGET_CUDA_FP16)
{
l1 = 0.0004;
lInf = 0.0012;
}
Mat outRef = readTorchBlob(_tf("net_openface_output.dat"), true);
normAssert(out, outRef, "", l1, lInf);
}
@ -343,6 +389,8 @@ TEST_P(Test_Torch_nets, ENet_accuracy)
checkBackend();
if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16)
throw SkipTestException("");
if (backend == DNN_BACKEND_CUDA && target == DNN_TARGET_CUDA_FP16)
applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA_FP16);
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target != DNN_TARGET_CPU)
{
if (target == DNN_TARGET_OPENCL_FP16) applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
@ -448,6 +496,10 @@ TEST_P(Test_Torch_nets, FastNeuralStyle_accuracy)
else
EXPECT_LE(normL1, 0.6f);
}
else if(target == DNN_TARGET_CUDA_FP16)
{
normAssert(out, refBlob, "", 0.6, 25);
}
else
normAssert(out, refBlob, "", 0.5, 1.1);
}

Write Preview
Loading…
Cancel
Save