potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-30 07:43:03 +04:00
Code Packages Releases Wiki Activity

Enable more deep learning tests

Browse Source
This commit is contained in:
Dmitry Kurtaev
2018-06-27 16:34:36 +03:00
parent a0baae8a55
commit 019c2f2115
13 changed files with 761 additions and 518 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/dnn/test/test_halide_layers.cpp
+206 -122
View File
@@ -12,32 +12,60 @@
namespace opencv_test { namespace {
#ifdef HAVE_HALIDE
using namespace cv;
using namespace cv::dnn;
using namespace testing;
static void test(LayerParams& params, Mat& input)
static void test(Mat& input, Net& net, int backendId, int targetId)
{
DNNTestLayer::checkBackend(backendId, targetId);
randu(input, -1.0f, 1.0f);
Net net;
int lid = net.addLayer(params.name, params.type, params);
net.connect(0, 0, lid, 0);
net.setInput(input);
net.setPreferableBackend(DNN_BACKEND_OPENCV);
Mat outputDefault = net.forward(params.name).clone();
Mat outputDefault = net.forward().clone();
net.setPreferableBackend(DNN_BACKEND_HALIDE);
Mat outputHalide = net.forward(params.name).clone();
normAssert(outputDefault, outputHalide);
net.setPreferableBackend(backendId);
net.setPreferableTarget(targetId);
Mat outputHalide = net.forward().clone();
double l1, lInf;
DNNTestLayer::getDefaultThresholds(backendId, targetId, &l1, &lInf);
normAssert(outputDefault, outputHalide, "", l1, lInf);
}
static void test(LayerParams& params, Mat& input, int backendId, int targetId)
{
Net net;
net.addLayerToPrev(params.name, params.type, params);
test(input, net, backendId, targetId);
}
static testing::internal::ParamGenerator<tuple<DNNBackend, DNNTarget> > dnnBackendsAndTargetsWithHalide()
{
static const tuple<DNNBackend, DNNTarget> testCases[] = {
#ifdef HAVE_HALIDE
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_HALIDE, DNN_TARGET_CPU),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_HALIDE, DNN_TARGET_OPENCL),
#endif
#ifdef HAVE_INF_ENGINE
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_CPU),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_OPENCL),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_OPENCL_FP16),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_MYRIAD),
#endif
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_OPENCV, DNN_TARGET_OPENCL),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_OPENCV, DNN_TARGET_OPENCL_FP16)
};
return testing::ValuesIn(testCases);
}
class Test_Halide_layers : public DNNTestLayer {};
////////////////////////////////////////////////////////////////////////////////
// Padding
////////////////////////////////////////////////////////////////////////////////
TEST(Padding_Halide, Accuracy)
TEST_P(Test_Halide_layers, Padding)
{
static const int kNumRuns = 10;
std::vector<int> paddings(8);
@@ -52,15 +80,16 @@ TEST(Padding_Halide, Accuracy)
lp.type = "Padding";
lp.name = "testLayer";
Mat input({1 + rng(10), 1 + rng(10), 1 + rng(10), 1 + rng(10)}, CV_32F);
test(lp, input);
int sz[] = {1 + (int)rng(10), 1 + (int)rng(10), 1 + (int)rng(10), 1 + (int)rng(10)};
Mat input(4, &sz[0], CV_32F);
test(lp, input, backend, target);
}
}
////////////////////////////////////////////////////////////////////////////////
// Convolution
////////////////////////////////////////////////////////////////////////////////
typedef TestWithParam<tuple<Vec3i, Size, Size, Size, Size, Size, bool> > Convolution;
typedef TestWithParam<tuple<Vec3i, Size, Size, Size, Size, Size, bool, tuple<DNNBackend, DNNTarget> > > Convolution;
TEST_P(Convolution, Accuracy)
{
int inChannels = get<0>(GetParam())[0];
@@ -72,8 +101,15 @@ TEST_P(Convolution, Accuracy)
Size pad = get<4>(GetParam());
Size dilation = get<5>(GetParam());
bool hasBias = get<6>(GetParam());
int backendId = get<0>(get<7>(GetParam()));
int targetId = get<1>(get<7>(GetParam()));
Mat weights({outChannels, inChannels / group, kernel.height, kernel.width}, CV_32F);
if ((backendId == DNN_BACKEND_INFERENCE_ENGINE && targetId == DNN_TARGET_MYRIAD) ||
(backendId == DNN_BACKEND_OPENCV && targetId == DNN_TARGET_OPENCL_FP16))
throw SkipTestException("");
int sz[] = {outChannels, inChannels / group, kernel.height, kernel.width};
Mat weights(4, &sz[0], CV_32F);
randu(weights, -1.0f, 1.0f);
LayerParams lp;
@@ -93,12 +129,13 @@ TEST_P(Convolution, Accuracy)
lp.blobs.push_back(weights);
if (hasBias)
{
Mat bias({outChannels}, CV_32F);
Mat bias(1, outChannels, CV_32F);
randu(bias, -1.0f, 1.0f);
lp.blobs.push_back(bias);
}
Mat input({1, inChannels, inSize.height, inSize.width}, CV_32F);
test(lp, input);
int inpSz[] = {1, inChannels, inSize.height, inSize.width};
Mat input(4, &inpSz[0], CV_32F);
test(lp, input, backendId, targetId);
}
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, Convolution, Combine(
@@ -110,13 +147,14 @@ INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, Convolution, Combine(
/*stride*/ Values(Size(1, 1), Size(2, 2)),
/*pad*/ Values(Size(1, 0), Size(0, 1)),
/*dilation*/ Values(Size(1, 1), Size(2, 2)),
/*has bias*/ Bool()
/*has bias*/ Bool(),
dnnBackendsAndTargetsWithHalide()
));
////////////////////////////////////////////////////////////////////////////////
// Deconvolution
////////////////////////////////////////////////////////////////////////////////
typedef TestWithParam<tuple<Vec3i, Size, Size, Size, Size, Vec4i, bool> > Deconvolution;
typedef TestWithParam<tuple<Vec3i, Size, Size, Size, Size, Vec4i, bool, tuple<DNNBackend, DNNTarget> > > Deconvolution;
TEST_P(Deconvolution, Accuracy)
{
int inChannels = get<0>(GetParam())[0];
@@ -129,8 +167,14 @@ TEST_P(Deconvolution, Accuracy)
Size stride = Size(get<5>(GetParam())[0], get<5>(GetParam())[1]);
Size adjPad = Size(get<5>(GetParam())[2], get<5>(GetParam())[3]);
bool hasBias = get<6>(GetParam());
int backendId = get<0>(get<7>(GetParam()));
int targetId = get<1>(get<7>(GetParam()));
if (backendId == DNN_BACKEND_INFERENCE_ENGINE && targetId == DNN_TARGET_CPU &&
dilation.width == 2 && dilation.height == 2)
throw SkipTestException("");
Mat weights({inChannels, outChannels / group, kernel.height, kernel.width}, CV_32F);
int sz[] = {inChannels, outChannels / group, kernel.height, kernel.width};
Mat weights(4, &sz[0], CV_32F);
randu(weights, -1.0f, 1.0f);
LayerParams lp;
@@ -152,12 +196,13 @@ TEST_P(Deconvolution, Accuracy)
lp.blobs.push_back(weights);
if (hasBias)
{
Mat bias({outChannels}, CV_32F);
Mat bias(1, outChannels, CV_32F);
randu(bias, -1.0f, 1.0f);
lp.blobs.push_back(bias);
}
Mat input({1, inChannels, inSize.height, inSize.width}, CV_32F);
test(lp, input);
int inpSz[] = {1, inChannels, inSize.height, inSize.width};
Mat input(4, &inpSz[0], CV_32F);
test(lp, input, backendId, targetId);
}
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, Deconvolution, Combine(
@@ -168,13 +213,14 @@ INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, Deconvolution, Combine(
/*pad*/ Values(Size(1, 0), Size(0, 1)),
/*dilation*/ Values(Size(1, 1), Size(2, 2)),
/*stride, adj. pad*/ Values(Vec4i(1,1, 0,0), Vec4i(2,2, 1,0), Vec4i(1,2, 0,1)),
/*has bias*/ Bool()
/*has bias*/ Bool(),
dnnBackendsAndTargetsWithHalide()
));
////////////////////////////////////////////////////////////////////////////////
// LRN
////////////////////////////////////////////////////////////////////////////////
typedef TestWithParam<tuple<Vec3i, int, Vec3f, bool, std::string> > LRN;
typedef TestWithParam<tuple<Vec3i, int, Vec3f, bool, std::string, tuple<DNNBackend, DNNTarget> > > LRN;
TEST_P(LRN, Accuracy)
{
int inChannels = get<0>(GetParam())[0];
@@ -185,6 +231,10 @@ TEST_P(LRN, Accuracy)
float bias = get<2>(GetParam())[2];
bool normBySize = get<3>(GetParam());
std::string nrmType = get<4>(GetParam());
int backendId = get<0>(get<5>(GetParam()));
int targetId = get<1>(get<5>(GetParam()));
if (backendId == DNN_BACKEND_INFERENCE_ENGINE)
throw SkipTestException("");
LayerParams lp;
lp.set("norm_region", nrmType);
@@ -196,8 +246,9 @@ TEST_P(LRN, Accuracy)
lp.type = "LRN";
lp.name = "testLayer";
Mat input({1, inChannels, inSize.height, inSize.width}, CV_32F);
test(lp, input);
int sz[] = {1, inChannels, inSize.height, inSize.width};
Mat input(4, &sz[0], CV_32F);
test(lp, input, backendId, targetId);
}
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, LRN, Combine(
@@ -207,19 +258,24 @@ INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, LRN, Combine(
/*alpha, beta,*/ Vec3f(1.0f, 0.9f, 1.1f), Vec3f(1.0f, 1.1f, 0.9f),
/*bias */ Vec3f(1.1f, 0.9f, 1.0f), Vec3f(1.1f, 1.0f, 0.9f)),
/*norm_by_size*/ Bool(),
/*norm_type*/ Values("ACROSS_CHANNELS", "WITHIN_CHANNEL")
/*norm_type*/ Values("ACROSS_CHANNELS", "WITHIN_CHANNEL"),
dnnBackendsAndTargetsWithHalide()
));
////////////////////////////////////////////////////////////////////////////////
// Average pooling
////////////////////////////////////////////////////////////////////////////////
typedef TestWithParam<tuple<int, Size, Size, Size> > AvePooling;
typedef TestWithParam<tuple<int, Size, Size, Size, tuple<DNNBackend, DNNTarget> > > AvePooling;
TEST_P(AvePooling, Accuracy)
{
int inChannels = get<0>(GetParam());
Size outSize = get<1>(GetParam());; // Input size will be computed from parameters.
Size kernel = get<2>(GetParam());
Size stride = get<3>(GetParam());
int backendId = get<0>(get<4>(GetParam()));
int targetId = get<1>(get<4>(GetParam()));
if (backendId == DNN_BACKEND_INFERENCE_ENGINE && targetId == DNN_TARGET_MYRIAD)
throw SkipTestException("");
const int inWidth = (outSize.width - 1) * stride.width + kernel.width;
const int inHeight = (outSize.height - 1) * stride.height + kernel.height;
@@ -233,21 +289,23 @@ TEST_P(AvePooling, Accuracy)
lp.type = "Pooling";
lp.name = "testLayer";
Mat input({1, inChannels, inHeight, inWidth}, CV_32F);
test(lp, input);
int sz[] = {1, inChannels, inHeight, inWidth};
Mat input(4, &sz[0], CV_32F);
test(lp, input, backendId, targetId);
}
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, AvePooling, Combine(
/*in channels*/ Values(3, 4),
/*out size*/ Values(Size(1, 1), Size(2, 2), Size(3, 2), Size(4, 7)),
/*kernel*/ Values(Size(1, 1), Size(2, 2), Size(3, 3), Size(3, 2)),
/*stride*/ Values(Size(1, 1), Size(2, 2), Size(3, 2))
/*stride*/ Values(Size(1, 1), Size(2, 2), Size(3, 2)),
dnnBackendsAndTargetsWithHalide()
));
////////////////////////////////////////////////////////////////////////////////
// Maximum pooling
////////////////////////////////////////////////////////////////////////////////
typedef TestWithParam<tuple<int, Size, Size, Size, Size> > MaxPooling;
typedef TestWithParam<tuple<int, Size, Size, Size, Size, tuple<DNNBackend, DNNTarget> > > MaxPooling;
TEST_P(MaxPooling, Accuracy)
{
int inChannels = get<0>(GetParam());
@@ -255,6 +313,8 @@ TEST_P(MaxPooling, Accuracy)
Size kernel = get<2>(GetParam());
Size stride = get<3>(GetParam());
Size pad = get<4>(GetParam());
int backendId = get<0>(get<5>(GetParam()));
int targetId = get<1>(get<5>(GetParam()));
LayerParams lp;
lp.set("pool", "max");
@@ -267,8 +327,9 @@ TEST_P(MaxPooling, Accuracy)
lp.type = "Pooling";
lp.name = "testLayer";
Mat input({1, inChannels, inSize.height, inSize.width}, CV_32F);
test(lp, input);
int sz[] = {1, inChannels, inSize.height, inSize.width};
Mat input(4, &sz[0], CV_32F);
test(lp, input, backendId, targetId);
}
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, MaxPooling, Combine(
@@ -276,19 +337,25 @@ INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, MaxPooling, Combine(
/*in size*/ Values(Size(5, 5), Size(7, 6)),
/*kernel*/ Values(Size(2, 2), Size(3, 3), Size(3, 2)),
/*stride*/ Values(Size(1, 1), Size(2, 2), Size(3, 2)),
/*pad*/ Values(Size(0, 0), Size(1, 1), Size(0, 1))
/*pad*/ Values(Size(0, 0), Size(1, 1), Size(0, 1)),
dnnBackendsAndTargetsWithHalide()
));
////////////////////////////////////////////////////////////////////////////////
// Fully-connected
////////////////////////////////////////////////////////////////////////////////
typedef TestWithParam<tuple<int, Size, int, bool> > FullyConnected;
typedef TestWithParam<tuple<int, Size, int, bool, tuple<DNNBackend, DNNTarget> > > FullyConnected;
TEST_P(FullyConnected, Accuracy)
{
int inChannels = get<0>(GetParam());
Size inSize = get<1>(GetParam());
int outChannels = get<2>(GetParam());
bool hasBias = get<3>(GetParam());
int backendId = get<0>(get<4>(GetParam()));
int targetId = get<1>(get<4>(GetParam()));
if (backendId == DNN_BACKEND_INFERENCE_ENGINE ||
(backendId == DNN_BACKEND_OPENCV && targetId == DNN_TARGET_OPENCL_FP16))
throw SkipTestException("");
Mat weights(outChannels, inChannels * inSize.height * inSize.width, CV_32F);
randu(weights, -1.0f, 1.0f);
@@ -304,39 +371,50 @@ TEST_P(FullyConnected, Accuracy)
lp.type = "InnerProduct";
lp.name = "testLayer";
Mat input({1, inChannels, inSize.height, inSize.width}, CV_32F);
test(lp, input);
int sz[] = {1, inChannels, inSize.height, inSize.width};
Mat input(4, &sz[0], CV_32F);
test(lp, input, backendId, targetId);
}
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, FullyConnected, Combine(
/*in channels*/ Values(3, 4),
/*in size*/ Values(Size(5, 4), Size(4, 5), Size(1, 1)),
/*out channels*/ Values(3, 4),
/*has bias*/ Bool()
/*has bias*/ Bool(),
dnnBackendsAndTargetsWithHalide()
));
////////////////////////////////////////////////////////////////////////////////
// SoftMax
////////////////////////////////////////////////////////////////////////////////
typedef TestWithParam<tuple<int> > SoftMax;
typedef TestWithParam<tuple<int, tuple<DNNBackend, DNNTarget> > > SoftMax;
TEST_P(SoftMax, Accuracy)
{
int inChannels = get<0>(GetParam());
int backendId = get<0>(get<1>(GetParam()));
int targetId = get<1>(get<1>(GetParam()));
LayerParams lp;
lp.type = "SoftMax";
lp.name = "testLayer";
Mat input({1, inChannels, 1, 1}, CV_32F);
test(lp, input);
int sz[] = {1, inChannels, 1, 1};
Mat input(4, &sz[0], CV_32F);
test(lp, input, backendId, targetId);
}
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, SoftMax, Values(3, 4, 5, 1024));
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, SoftMax, Combine(
Values(3, 4, 5, 1024),
dnnBackendsAndTargetsWithHalide()
));
//////////////////////////////////////////////////////////////////////////////
// Max pooling - unpooling
//////////////////////////////////////////////////////////////////////////////
TEST(MaxPoolUnpool_Halide, Accuracy)
TEST_P(Test_Halide_layers, MaxPoolUnpool)
{
if (backend == DNN_BACKEND_INFERENCE_ENGINE)
throw SkipTestException("");
LayerParams pool;
pool.set("pool", "max");
pool.set("kernel_w", 2);
@@ -366,16 +444,9 @@ TEST(MaxPoolUnpool_Halide, Accuracy)
net.connect(poolId, 0, unpoolId, 0);
net.connect(poolId, 1, unpoolId, 1);
Mat input({1, 1, 4, 4}, CV_32F);
randu(input, -1.0f, 1.0f);
net.setInput(input);
net.setPreferableBackend(DNN_BACKEND_OPENCV);
Mat outputDefault = net.forward("testUnpool").clone();
net.setPreferableBackend(DNN_BACKEND_HALIDE);
net.setInput(input);
Mat outputHalide = net.forward("testUnpool").clone();
normAssert(outputDefault, outputHalide);
int sz[] = {1, 1, 4, 4};
Mat input(4, &sz[0], CV_32F);
test(input, net, backend, target);
}
////////////////////////////////////////////////////////////////////////////////
@@ -383,7 +454,7 @@ TEST(MaxPoolUnpool_Halide, Accuracy)
////////////////////////////////////////////////////////////////////////////////
static const int kNumChannels = 3;
void testInPlaceActivation(LayerParams& lp)
void testInPlaceActivation(LayerParams& lp, int backendId, int targetId)
{
EXPECT_FALSE(lp.name.empty());
@@ -400,24 +471,19 @@ void testInPlaceActivation(LayerParams& lp)
net.connect(0, 0, poolId, 0);
net.addLayerToPrev(lp.name, lp.type, lp);
Mat input({1, kNumChannels, 10, 10}, CV_32F);
randu(input, -1.0f, 1.0f);
net.setInput(input);
net.setPreferableBackend(DNN_BACKEND_OPENCV);
Mat outputDefault = net.forward(lp.name).clone();
net.setInput(input);
net.setPreferableBackend(DNN_BACKEND_HALIDE);
Mat outputHalide = net.forward(lp.name).clone();
normAssert(outputDefault, outputHalide);
int sz[] = {1, kNumChannels, 10, 10};
Mat input(4, &sz[0], CV_32F);
test(input, net, backendId, targetId);
}
typedef TestWithParam<tuple<bool, bool, float> > BatchNorm;
typedef TestWithParam<tuple<bool, bool, float, tuple<DNNBackend, DNNTarget> > > BatchNorm;
TEST_P(BatchNorm, Accuracy)
{
bool hasWeights = get<0>(GetParam());
bool hasBias = get<1>(GetParam());
float epsilon = get<2>(GetParam());
int backendId = get<0>(get<3>(GetParam()));
int targetId = get<1>(get<3>(GetParam()));
LayerParams lp;
lp.set("has_weight", hasWeights);
@@ -428,56 +494,66 @@ TEST_P(BatchNorm, Accuracy)
lp.blobs.reserve(4);
for (int i = 0; i < 3; ++i)
lp.blobs.push_back(Mat({kNumChannels}, CV_32F));
lp.blobs.push_back(Mat(1, kNumChannels, CV_32F));
if (hasBias || hasWeights)
lp.blobs.push_back(Mat({kNumChannels}, CV_32F));
lp.blobs.push_back(Mat(1, kNumChannels, CV_32F));
for (Mat& m : lp.blobs)
randu(m, 0.0f, 1.0f);
for (int i = 0; i < lp.blobs.size(); ++i)
randu(lp.blobs[i], 0.0f, 1.0f);
testInPlaceActivation(lp);
testInPlaceActivation(lp, backendId, targetId);
}
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, BatchNorm, Combine(
/*has weights*/ Bool(),
/*has bias*/ Bool(),
/*epsilon*/ Values(1e-3f, 1e-5f)
/*epsilon*/ Values(1e-3f, 1e-5f),
dnnBackendsAndTargetsWithHalide()
));
typedef TestWithParam<tuple<float> > ReLU;
typedef TestWithParam<tuple<float, tuple<DNNBackend, DNNTarget> > > ReLU;
TEST_P(ReLU, Accuracy)
{
float negativeSlope = get<0>(GetParam());
int backendId = get<0>(get<1>(GetParam()));
int targetId = get<1>(get<1>(GetParam()));
LayerParams lp;
lp.set("negative_slope", negativeSlope);
lp.type = "ReLU";
lp.name = "testLayer";
testInPlaceActivation(lp);
testInPlaceActivation(lp, backendId, targetId);
}
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, ReLU, Values(
/*negative slope*/ 2.0f, 0.3f, -0.1f, 0.0f
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, ReLU, Combine(
/*negative slope*/ Values(2.0f, 0.3f, -0.1f, 0.0f),
dnnBackendsAndTargetsWithHalide()
));
typedef TestWithParam<tuple<std::string> > NoParamActivation;
typedef TestWithParam<tuple<std::string, tuple<DNNBackend, DNNTarget> > > NoParamActivation;
TEST_P(NoParamActivation, Accuracy)
{
int backendId = get<0>(get<1>(GetParam()));
int targetId = get<1>(get<1>(GetParam()));
LayerParams lp;
lp.type = get<0>(GetParam());
lp.name = "testLayer";
testInPlaceActivation(lp);
testInPlaceActivation(lp, backendId, targetId);
}
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, NoParamActivation, Values(
/*type*/ "TanH", "Sigmoid", "AbsVal", "BNLL"
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, NoParamActivation, Combine(
/*type*/ Values("TanH", "Sigmoid", "AbsVal", "BNLL"),
dnnBackendsAndTargetsWithHalide()
));
typedef TestWithParam<tuple<Vec3f> > Power;
typedef TestWithParam<tuple<Vec3f, tuple<DNNBackend, DNNTarget> > > Power;
TEST_P(Power, Accuracy)
{
float power = get<0>(GetParam())[0];
float scale = get<0>(GetParam())[1];
float shift = get<0>(GetParam())[2];
int backendId = get<0>(get<1>(GetParam()));
int targetId = get<1>(get<1>(GetParam()));
LayerParams lp;
lp.set("power", power);
@@ -485,46 +561,52 @@ TEST_P(Power, Accuracy)
lp.set("shift", shift);
lp.type = "Power";
lp.name = "testLayer";
testInPlaceActivation(lp);
testInPlaceActivation(lp, backendId, targetId);
}
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, Power,
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, Power, Combine(
/*power, scale, shift*/ Values(Vec3f(0.9f, 1.0f, 1.1f), Vec3f(0.9f, 1.1f, 1.0f),
Vec3f(1.0f, 0.9f, 1.1f), Vec3f(1.0f, 1.1f, 0.9f),
Vec3f(1.1f, 0.9f, 1.0f), Vec3f(1.1f, 1.0f, 0.9f))
);
Vec3f(1.1f, 0.9f, 1.0f), Vec3f(1.1f, 1.0f, 0.9f)),
dnnBackendsAndTargetsWithHalide()
));
TEST(ChannelsPReLU, Accuracy)
TEST_P(Test_Halide_layers, ChannelsPReLU)
{
LayerParams lp;
lp.type = "ChannelsPReLU";
lp.name = "testLayer";
lp.blobs.push_back(Mat({kNumChannels}, CV_32F));
lp.blobs.push_back(Mat(1, kNumChannels, CV_32F));
randu(lp.blobs[0], -1.0f, 1.0f);
testInPlaceActivation(lp);
testInPlaceActivation(lp, backend, target);
}
typedef TestWithParam<tuple<bool> > Scale;
typedef TestWithParam<tuple<bool, tuple<DNNBackend, DNNTarget> > > Scale;
TEST_P(Scale, Accuracy)
{
bool hasBias = get<0>(GetParam());
int backendId = get<0>(get<1>(GetParam()));
int targetId = get<1>(get<1>(GetParam()));
LayerParams lp;
lp.set("bias_term", hasBias);
lp.type = "Scale";
lp.name = "testLayer";
lp.blobs.push_back(Mat({kNumChannels}, CV_32F));
lp.blobs.push_back(Mat(1, kNumChannels, CV_32F));
randu(lp.blobs[0], -1.0f, 1.0f);
if (hasBias)
{
lp.blobs.push_back(Mat({kNumChannels}, CV_32F));
lp.blobs.push_back(Mat(1, kNumChannels, CV_32F));
randu(lp.blobs[1], -1.0f, 1.0f);
}
testInPlaceActivation(lp);
testInPlaceActivation(lp, backendId, targetId);
}
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, Scale, Values(true, false));
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, Scale, Combine(
Bool(),
dnnBackendsAndTargetsWithHalide()
));
////////////////////////////////////////////////////////////////////////////////
// Concat layer
@@ -534,11 +616,13 @@ INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, Scale, Values(true, false));
// `--- conv ----^ ^ ^
// `---- ... ------' '
// `-----------------'
typedef TestWithParam<tuple<Vec3i, Vec3i> > Concat;
typedef TestWithParam<tuple<Vec3i, Vec3i, tuple<DNNBackend, DNNTarget> > > Concat;
TEST_P(Concat, Accuracy)
{
Vec3i inSize = get<0>(GetParam());
Vec3i numChannels = get<1>(GetParam());
int backendId = get<0>(get<2>(GetParam()));
int targetId = get<1>(get<2>(GetParam()));
Net net;
@@ -549,7 +633,8 @@ TEST_P(Concat, Accuracy)
if (!numChannels[i])
break;
Mat weights({numChannels[i], inSize[0], 1, 1}, CV_32F);
int sz[] = {numChannels[i], inSize[0], 1, 1};
Mat weights(4, &sz[0], CV_32F);
randu(weights, -1.0f, 1.0f);
LayerParams convParam;
@@ -578,21 +663,15 @@ TEST_P(Concat, Accuracy)
net.connect(convLayerIds[i], 0, concatId, i + 1);
}
Mat input({1, inSize[0], inSize[1], inSize[2]}, CV_32F);
randu(input, -1.0f, 1.0f);
net.setInput(input);
net.setPreferableBackend(DNN_BACKEND_OPENCV);
Mat outputDefault = net.forward(concatParam.name).clone();
net.setPreferableBackend(DNN_BACKEND_HALIDE);
Mat outputHalide = net.forward(concatParam.name).clone();
normAssert(outputDefault, outputHalide);
int sz[] = {1, inSize[0], inSize[1], inSize[2]};
Mat input(4, &sz[0], CV_32F);
test(input, net, backendId, targetId);
}
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, Concat, Combine(
/*input size*/ Values(Vec3i(1, 4, 5), Vec3i(2, 8, 6)),
/*channels*/ Values(Vec3i(2, 0, 0), Vec3i(3, 4, 0), Vec3i(1, 6, 2))
/*channels*/ Values(Vec3i(2, 0, 0), Vec3i(3, 4, 0), Vec3i(1, 6, 2)),
dnnBackendsAndTargetsWithHalide()
));
////////////////////////////////////////////////////////////////////////////////
@@ -603,20 +682,27 @@ INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, Concat, Combine(
// `--- conv ----^ ^ ^
// `---- ... ------' '
// `-----------------'
typedef TestWithParam<tuple<Vec3i, std::string, int, bool> > Eltwise;
typedef TestWithParam<tuple<Vec3i, std::string, int, bool, tuple<DNNBackend, DNNTarget> > > Eltwise;
TEST_P(Eltwise, Accuracy)
{
Vec3i inSize = get<0>(GetParam());
std::string op = get<1>(GetParam());
int numConv = get<2>(GetParam());
bool weighted = get<3>(GetParam());
int backendId = get<0>(get<4>(GetParam()));
int targetId = get<1>(get<4>(GetParam()));
if (backendId == DNN_BACKEND_OPENCV &&
(targetId == DNN_TARGET_OPENCL || targetId == DNN_TARGET_OPENCL_FP16))
throw SkipTestException("");
Net net;
std::vector<int> convLayerIds(numConv);
for (int i = 0; i < numConv; ++i)
{
Mat weights({inSize[0], inSize[0], 1, 1}, CV_32F);
int sz[] = {inSize[0], inSize[0], 1, 1};
Mat weights(4, &sz[0], CV_32F);
randu(weights, -1.0f, 1.0f);
LayerParams convParam;
@@ -655,28 +741,23 @@ TEST_P(Eltwise, Accuracy)
net.connect(convLayerIds[i], 0, eltwiseId, i + 1);
}
Mat input({1, inSize[0], inSize[1], inSize[2]}, CV_32F);
randu(input, -1.0f, 1.0f);
net.setInput(input);
net.setPreferableBackend(DNN_BACKEND_OPENCV);
Mat outputDefault = net.forward(eltwiseParam.name).clone();
net.setPreferableBackend(DNN_BACKEND_HALIDE);
Mat outputHalide = net.forward(eltwiseParam.name).clone();
normAssert(outputDefault, outputHalide);
int sz[] = {1, inSize[0], inSize[1], inSize[2]};
Mat input(4, &sz[0], CV_32F);
test(input, net, backendId, targetId);
}
INSTANTIATE_TEST_CASE_P(Layer_Test_Halide, Eltwise, Combine(
/*input size*/ Values(Vec3i(1, 4, 5), Vec3i(2, 8, 6)),
/*operation*/ Values("prod", "sum", "max"),
/*num convs*/ Values(1, 2, 3),
/*weighted(for sum only)*/ Bool()
/*weighted(for sum only)*/ Bool(),
dnnBackendsAndTargetsWithHalide()
));
////////////////////////////////////////////////////////////////////////////
// Mixed backends
////////////////////////////////////////////////////////////////////////////
#ifdef HAVE_HALIDE
TEST(MixedBackends_Halide_Default_Halide, Accuracy)
{
// Just a layer that supports Halide backend.
@@ -700,7 +781,8 @@ TEST(MixedBackends_Halide_Default_Halide, Accuracy)
net.addLayerToPrev(mvn.name, mvn.type, mvn);
net.addLayerToPrev(lrn2.name, lrn2.type, lrn2);
Mat input({4, 3, 5, 6}, CV_32F);
int sz[] = {4, 3, 5, 6};
Mat input(4, &sz[0], CV_32F);
randu(input, -1.0f, 1.0f);
net.setInput(input);
net.setPreferableBackend(DNN_BACKEND_OPENCV);
@@ -718,4 +800,6 @@ TEST(MixedBackends_Halide_Default_Halide, Accuracy)
}
#endif // HAVE_HALIDE
INSTANTIATE_TEST_CASE_P(/*nothing*/, Test_Halide_layers, dnnBackendsAndTargetsWithHalide());
}} // namespace