potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-30 15:53:03 +04:00
Code Packages Releases Wiki Activity

Merge remote-tracking branch 'upstream/3.4' into merge-3.4

Browse Source
This commit is contained in:
Alexander Alekhin
2020-05-06 19:52:58 +00:00
parent c722625f28 5813d2439c
commit 09799402f9
10 changed files with 229 additions and 67 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/dnn/src/darknet/darknet_io.cpp
+14 -4
View File
@@ -229,6 +229,10 @@ namespace cv {
{
activation_param.type = "Swish";
}
else if (type == "mish")
{
activation_param.type = "Mish";
}
else if (type == "logistic")
{
activation_param.type = "Sigmoid";
@@ -436,7 +440,7 @@ namespace cv {
fused_layer_names.push_back(last_layer);
}
void setYolo(int classes, const std::vector<int>& mask, const std::vector<float>& anchors, float thresh, float nms_threshold)
void setYolo(int classes, const std::vector<int>& mask, const std::vector<float>& anchors, float thresh, float nms_threshold, float scale_x_y)
{
cv::dnn::LayerParams region_param;
region_param.name = "Region-name";
@@ -449,6 +453,7 @@ namespace cv {
region_param.set<bool>("logistic", true);
region_param.set<float>("thresh", thresh);
region_param.set<float>("nms_threshold", nms_threshold);
region_param.set<float>("scale_x_y", scale_x_y);
std::vector<float> usedAnchors(numAnchors * 2);
for (int i = 0; i < numAnchors; ++i)
@@ -786,6 +791,7 @@ namespace cv {
int num_of_anchors = getParam<int>(layer_params, "num", -1);
float thresh = getParam<float>(layer_params, "thresh", 0.2);
float nms_threshold = getParam<float>(layer_params, "nms_threshold", 0.4);
float scale_x_y = getParam<float>(layer_params, "scale_x_y", 1.0);
std::string anchors_values = getParam<std::string>(layer_params, "anchors", std::string());
CV_Assert(!anchors_values.empty());
@@ -798,7 +804,7 @@ namespace cv {
CV_Assert(classes > 0 && num_of_anchors > 0 && (num_of_anchors * 2) == anchors_vec.size());
setParams.setPermute(false);
setParams.setYolo(classes, mask_vec, anchors_vec, thresh, nms_threshold);
setParams.setYolo(classes, mask_vec, anchors_vec, thresh, nms_threshold, scale_x_y);
}
else {
CV_Error(cv::Error::StsParseError, "Unknown layer type: " + layer_type);
@@ -813,6 +819,10 @@ namespace cv {
{
setParams.setActivation("swish");
}
else if (activation == "mish")
{
setParams.setActivation("mish");
}
else if (activation == "logistic")
{
setParams.setActivation("logistic");
@@ -935,8 +945,8 @@ namespace cv {
}
std::string activation = getParam<std::string>(layer_params, "activation", "linear");
if(activation == "leaky" || activation == "swish" || activation == "logistic")
++cv_layers_counter; // For ReLU, Swish, Sigmoid
if(activation == "leaky" || activation == "swish" || activation == "mish" || activation == "logistic")
++cv_layers_counter; // For ReLU, Swish, Mish, Sigmoid
if(!darknet_layers_counter)
tensor_shape.resize(1);
modules/dnn/src/layers/region_layer.cpp
+14 -3
View File
@@ -69,7 +69,7 @@ class RegionLayerImpl CV_FINAL : public RegionLayer
{
public:
int coords, classes, anchors, classfix;
float thresh, nmsThreshold;
float thresh, nmsThreshold, scale_x_y;
bool useSoftmax, useLogistic;
#ifdef HAVE_OPENCL
UMat blob_umat;
@@ -88,6 +88,7 @@ public:
useSoftmax = params.get<bool>("softmax", false);
useLogistic = params.get<bool>("logistic", false);
nmsThreshold = params.get<float>("nms_threshold", 0.4);
scale_x_y = params.get<float>("scale_x_y", 1.0); // Yolov4
CV_Assert(nmsThreshold >= 0.);
CV_Assert(coords == 4);
@@ -302,8 +303,10 @@ public:
if (classfix == -1 && scale < .5) scale = 0; // if(t0 < 0.5) t0 = 0;
int box_index = index_sample_offset + index * cell_size;
dstData[box_index + 0] = (x + logistic_activate(srcData[box_index + 0])) / cols;
dstData[box_index + 1] = (y + logistic_activate(srcData[box_index + 1])) / rows;
float x_tmp = (logistic_activate(srcData[box_index + 0]) - 0.5f) * scale_x_y + 0.5f;
float y_tmp = (logistic_activate(srcData[box_index + 1]) - 0.5f) * scale_x_y + 0.5f;
dstData[box_index + 0] = (x + x_tmp) / cols;
dstData[box_index + 1] = (y + y_tmp) / rows;
dstData[box_index + 2] = exp(srcData[box_index + 2]) * biasData[2 * a] / wNorm;
dstData[box_index + 3] = exp(srcData[box_index + 3]) * biasData[2 * a + 1] / hNorm;
@@ -471,6 +474,8 @@ public:
auto shape_3d = std::make_shared<ngraph::op::Constant>(ngraph::element::i64, ngraph::Shape{boxes_shape.size()}, boxes_shape.data());
ngraph::Shape box_broad_shape{1, (size_t)anchors, (size_t)h, (size_t)w};
auto scale_x_y_node = std::make_shared<ngraph::op::Constant>(ngraph::element::f32, ngraph::Shape{1}, &scale_x_y);
auto shift_node = std::make_shared<ngraph::op::Constant>(ngraph::element::f32, ngraph::Shape{1}, std::vector<float>{0.5});
std::shared_ptr<ngraph::Node> box_x;
{
@@ -478,6 +483,9 @@ public:
auto upper_bounds = std::make_shared<ngraph::op::Constant>(ngraph::element::i64, ngraph::Shape{2}, std::vector<int64_t>{1, cols});
box_x = std::make_shared<ngraph::op::v1::StridedSlice>(input2d, lower_bounds, upper_bounds, strides, std::vector<int64_t>{}, std::vector<int64_t>{});
box_x = std::make_shared<ngraph::op::Sigmoid>(box_x);
box_x = std::make_shared<ngraph::op::v1::Subtract>(box_x, shift_node, ngraph::op::AutoBroadcastType::NUMPY);
box_x = std::make_shared<ngraph::op::v1::Multiply>(box_x, scale_x_y_node, ngraph::op::AutoBroadcastType::NUMPY);
box_x = std::make_shared<ngraph::op::v1::Add>(box_x, shift_node, ngraph::op::AutoBroadcastType::NUMPY);
box_x = std::make_shared<ngraph::op::v1::Reshape>(box_x, shape_3d, true);
std::vector<float> x_indices(w * h * anchors);
@@ -504,6 +512,9 @@ public:
auto upper_bounds = std::make_shared<ngraph::op::Constant>(ngraph::element::i64, ngraph::Shape{2}, std::vector<int64_t>{2, cols});
box_y = std::make_shared<ngraph::op::v1::StridedSlice>(input2d, lower_bounds, upper_bounds, strides, std::vector<int64_t>{}, std::vector<int64_t>{});
box_y = std::make_shared<ngraph::op::Sigmoid>(box_y);
box_y = std::make_shared<ngraph::op::v1::Subtract>(box_y, shift_node, ngraph::op::AutoBroadcastType::NUMPY);
box_y = std::make_shared<ngraph::op::v1::Multiply>(box_y, scale_x_y_node, ngraph::op::AutoBroadcastType::NUMPY);
box_y = std::make_shared<ngraph::op::v1::Add>(box_y, shift_node, ngraph::op::AutoBroadcastType::NUMPY);
box_y = std::make_shared<ngraph::op::v1::Reshape>(box_y, shape_3d, true);
std::vector<float> y_indices(h * anchors);
modules/dnn/src/layers/slice_layer.cpp
+32 -27
View File
@@ -172,18 +172,19 @@ public:
CV_Assert(inputs.size() == 1);
const MatSize& inpShape = inputs[0].size;
finalSliceRanges = sliceRanges;
if (sliceRanges.empty())
{
// Divide input blob on equal parts by axis.
int outAxisSize = inpShape[axis] / outputs.size();
sliceRanges.resize(outputs.size(),
std::vector<Range>(axis + 1, Range::all()));
finalSliceRanges.resize(outputs.size(),
std::vector<Range>(axis + 1, Range::all()));
int prevSlice = 0;
for (int i = 0; i < outputs.size(); ++i)
{
sliceRanges[i][axis].start = prevSlice;
sliceRanges[i][axis].end = sliceRanges[i][axis].start + outAxisSize;
prevSlice = sliceRanges[i][axis].end;
finalSliceRanges[i][axis].start = prevSlice;
finalSliceRanges[i][axis].end = finalSliceRanges[i][axis].start + outAxisSize;
prevSlice = finalSliceRanges[i][axis].end;
}
}
else
@@ -191,16 +192,16 @@ public:
for (int i = 0; i < outputs.size(); ++i)
{
CV_Assert(sliceRanges[i].size() <= inpShape.dims());
CV_Assert(finalSliceRanges[i].size() <= inpShape.dims());
// Fill the rest of ranges.
for (int j = sliceRanges[i].size(); j < inpShape.dims(); ++j)
for (int j = finalSliceRanges[i].size(); j < inpShape.dims(); ++j)
{
sliceRanges[i].push_back(Range::all());
finalSliceRanges[i].push_back(Range::all());
}
// Clamp.
for (int j = 0; j < sliceRanges[i].size(); ++j)
for (int j = 0; j < finalSliceRanges[i].size(); ++j)
{
sliceRanges[i][j] = clamp(sliceRanges[i][j], inpShape[j]);
finalSliceRanges[i][j] = clamp(finalSliceRanges[i][j], inpShape[j]);
}
}
}
@@ -241,8 +242,8 @@ public:
kernel.set(idx++, (int)(rows * cols));
kernel.set(idx++, (int)inpMat.size[3]);
kernel.set(idx++, (int)cols);
kernel.set(idx++, (int)sliceRanges[i][2].start);
kernel.set(idx++, (int)sliceRanges[i][3].start);
kernel.set(idx++, (int)finalSliceRanges[i][2].start);
kernel.set(idx++, (int)finalSliceRanges[i][3].start);
kernel.set(idx++, ocl::KernelArg::PtrWriteOnly(outputs[i]));
bool ret = kernel.run(1, global, local, false);
if (!ret)
@@ -266,10 +267,10 @@ public:
outputs_arr.getMatVector(outputs);
const Mat& inpMat = inputs[0];
CV_Assert(outputs.size() == sliceRanges.size());
CV_Assert(outputs.size() == finalSliceRanges.size());
for (size_t i = 0; i < outputs.size(); i++)
{
inpMat(sliceRanges[i]).copyTo(outputs[i]);
inpMat(finalSliceRanges[i]).copyTo(outputs[i]);
}
}
@@ -278,11 +279,11 @@ public:
#if INF_ENGINE_VER_MAJOR_GE(INF_ENGINE_RELEASE_2019R1)
virtual Ptr<BackendNode> initInfEngine(const std::vector<Ptr<BackendWrapper> >& inputs) CV_OVERRIDE
{
CV_Assert_N(sliceRanges.size() == 1, inputs.size() <= 2);
CV_Assert_N(finalSliceRanges.size() == 1, inputs.size() <= 2);
std::vector<size_t> axes, offsets, dims;
int from, to, step;
int numDims = sliceRanges[0].size();
int numDims = finalSliceRanges[0].size();
if (preferableTarget == DNN_TARGET_MYRIAD)
{
from = axis;
@@ -298,8 +299,8 @@ public:
for (int i = from; i != to; i += step)
{
axes.push_back(i);
offsets.push_back(sliceRanges[0][i].start);
dims.push_back(sliceRanges[0][i].size());
offsets.push_back(finalSliceRanges[0][i].start);
dims.push_back(finalSliceRanges[0][i].size());
}
InferenceEngine::Builder::Layer ieLayer(name);
@@ -315,7 +316,7 @@ public:
{
std::vector<size_t> outShape(numDims);
for (int i = 0; i < numDims; ++i)
outShape[i] = sliceRanges[0][i].size();
outShape[i] = finalSliceRanges[0][i].size();
ieLayer.getInputPorts()[1].setParameter("type", "weights");
@@ -338,13 +339,13 @@ public:
{
CV_Assert_N(nodes.size() <= 2);
auto& ieInpNode = nodes[0].dynamicCast<InfEngineNgraphNode>()->node;
CV_Assert(sliceRanges[0].size() == ieInpNode->get_shape().size());
CV_Assert(finalSliceRanges[0].size() == ieInpNode->get_shape().size());
std::vector<int64_t> offsets, dims;
for (int i = 0; i < sliceRanges[0].size(); ++i)
for (int i = 0; i < finalSliceRanges[0].size(); ++i)
{
offsets.push_back(sliceRanges[0][i].start);
dims.push_back(sliceRanges[0][i].end);
offsets.push_back(finalSliceRanges[0][i].start);
dims.push_back(finalSliceRanges[0][i].end);
}
auto lower_bounds = std::make_shared<ngraph::op::Constant>(ngraph::element::i64,
@@ -384,6 +385,10 @@ public:
}
#endif
protected:
// The actual non-negative values determined from @p sliceRanges depends on input size.
std::vector<std::vector<Range> > finalSliceRanges;
};
class CropLayerImpl CV_FINAL : public SliceLayerImpl
@@ -447,18 +452,18 @@ public:
offset_final[i] = offset[i - start_axis];
}
sliceRanges.resize(1);
sliceRanges[0].resize(dims);
finalSliceRanges.resize(1);
finalSliceRanges[0].resize(dims);
for (int i = 0; i < start_axis; i++)
{
sliceRanges[0][i] = Range(0, inpBlob.size[i]);
finalSliceRanges[0][i] = Range(0, inpBlob.size[i]);
}
for (int i = start_axis; i < dims; i++)
{
if (offset_final[i] < 0 || offset_final[i] + inpSzBlob.size[i] > inpBlob.size[i])
CV_Error(Error::StsBadArg, "invalid crop parameters or blob sizes");
sliceRanges[0][i] = Range(offset_final[i], offset_final[i] + inpSzBlob.size[i]);
finalSliceRanges[0][i] = Range(offset_final[i], offset_final[i] + inpSzBlob.size[i]);
}
}
modules/dnn/test/test_darknet_importer.cpp
+5
View File
@@ -549,6 +549,11 @@ TEST_P(Test_Darknet_layers, upsample)
testDarknetLayer("upsample");
}
TEST_P(Test_Darknet_layers, mish)
{
testDarknetLayer("mish", true);
}
TEST_P(Test_Darknet_layers, avgpool_softmax)
{
testDarknetLayer("avgpool_softmax");
modules/dnn/test/test_layers.cpp
+34
View File
@@ -1791,4 +1791,38 @@ TEST_P(Layer_Test_Resize, change_input)
INSTANTIATE_TEST_CASE_P(/**/, Layer_Test_Resize, dnnBackendsAndTargets());
typedef testing::TestWithParam<tuple<Backend, Target> > Layer_Test_Slice;
TEST_P(Layer_Test_Slice, variable_input_shape)
{
int backendId = get<0>(GetParam());
int targetId = get<1>(GetParam());
int begin[] = {0, 0, 0, 0};
int end[] = {-1, -1, -1, -1};
Net net;
LayerParams lp;
lp.type = "Slice";
lp.name = "testLayer";
lp.set("begin", DictValue::arrayInt<int*>(&begin[0], 4));
lp.set("end", DictValue::arrayInt<int*>(&end[0], 4));
net.addLayerToPrev(lp.name, lp.type, lp);
for (int i = 0; i < 2; ++i)
{
Mat inp(4 + i, 5 + i, CV_8UC1);
randu(inp, 0, 255);
inp = blobFromImage(inp);
net.setInput(inp);
net.setPreferableBackend(backendId);
net.setPreferableTarget(targetId);
Mat out = net.forward();
normAssert(out, inp);
}
}
INSTANTIATE_TEST_CASE_P(/**/, Layer_Test_Slice, dnnBackendsAndTargets());
}} // namespace
modules/dnn/test/test_onnx_importer.cpp
+3 -3
View File
@@ -559,7 +559,7 @@ public:
TEST_P(Test_ONNX_nets, Alexnet)
{
#if defined(OPENCV_32BIT_CONFIGURATION) && defined(HAVE_OPENCL)
#if defined(OPENCV_32BIT_CONFIGURATION) && (defined(HAVE_OPENCL) || defined(_WIN32))
applyTestTag(CV_TEST_TAG_MEMORY_2GB);
#else
applyTestTag(target == DNN_TARGET_CPU ? CV_TEST_TAG_MEMORY_512MB : CV_TEST_TAG_MEMORY_1GB);
@@ -623,7 +623,7 @@ TEST_P(Test_ONNX_nets, Googlenet)
TEST_P(Test_ONNX_nets, CaffeNet)
{
#if defined(OPENCV_32BIT_CONFIGURATION) && defined(HAVE_OPENCL)
#if defined(OPENCV_32BIT_CONFIGURATION) && (defined(HAVE_OPENCL) || defined(_WIN32))
applyTestTag(CV_TEST_TAG_MEMORY_2GB);
#else
applyTestTag(target == DNN_TARGET_CPU ? CV_TEST_TAG_MEMORY_512MB : CV_TEST_TAG_MEMORY_1GB);
@@ -639,7 +639,7 @@ TEST_P(Test_ONNX_nets, CaffeNet)
TEST_P(Test_ONNX_nets, RCNN_ILSVRC13)
{
#if defined(OPENCV_32BIT_CONFIGURATION) && defined(HAVE_OPENCL)
#if defined(OPENCV_32BIT_CONFIGURATION) && (defined(HAVE_OPENCL) || defined(_WIN32))
applyTestTag(CV_TEST_TAG_MEMORY_2GB);
#else
applyTestTag(target == DNN_TARGET_CPU ? CV_TEST_TAG_MEMORY_512MB : CV_TEST_TAG_MEMORY_1GB);