potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-29 07:13:02 +04:00
Code Packages Releases Wiki Activity

Merge pull request #12264 from dkurt:dnn_remove_forward_method

Browse Source 
* Remove a forward method in dnn::Layer

* Add a test

* Fix tests

* Mark multiple dnn::Layer::finalize methods as deprecated

* Replace back dnn's inputBlobs to vector of pointers

* Remove Layer::forward_fallback from CV_OCL_RUN scopes
This commit is contained in:
Dmitry Kurtaev
2018-09-06 13:26:47 +03:00
committed by Alexander Alekhin
parent 8e8315abfd
commit d486204a0d
40 changed files with 820 additions and 487 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/dnn/src/layers/convolution_layer.cpp
+74 -59
View File
@@ -79,49 +79,24 @@ public:
adjustPad.height < stride.height);
}
virtual bool supportBackend(int backendId) CV_OVERRIDE
void finalize(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr) CV_OVERRIDE
{
#ifdef HAVE_INF_ENGINE
if (backendId == DNN_BACKEND_INFERENCE_ENGINE)
{
if (type == "Convolution")
return preferableTarget != DNN_TARGET_MYRIAD || dilation.width == dilation.height;
else
{
CV_Assert(type == "Deconvolution");
const int outGroupCn = blobs[0].size[1]; // Weights are in IOHW layout
const int group = numOutput / outGroupCn;
if (group != 1)
{
#if INF_ENGINE_VER_MAJOR_GE(INF_ENGINE_RELEASE_2018R3)
return preferableTarget == DNN_TARGET_CPU;
#endif
return false;
}
if (preferableTarget == DNN_TARGET_OPENCL || preferableTarget == DNN_TARGET_OPENCL_FP16)
return dilation.width == 1 && dilation.height == 1;
return true;
}
}
else
#endif // HAVE_INF_ENGINE
return backendId == DNN_BACKEND_OPENCV || backendId == DNN_BACKEND_HALIDE;
}
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
void finalize(const std::vector<Mat*> &inputs, std::vector<Mat> &outputs) CV_OVERRIDE
{
CV_Assert(inputs.size() > 0);
CV_Assert(blobs.size() >= 1 && blobs.size() <= 2);
CV_Assert(blobs[0].dims == 4 && blobs[0].size[3] == kernel.width && blobs[0].size[2] == kernel.height);
const Mat &input = *inputs[0];
const Mat &input = inputs[0];
CV_Assert(input.dims == 4 && (input.type() == CV_32F || input.type() == CV_64F || input.type() == CV_16S));
for (size_t i = 0; i < inputs.size(); i++)
{
CV_Assert(inputs[i]->type() == input.type());
CV_Assert(inputs[i]->dims == 4 && inputs[i]->size[1] == input.size[1]);
CV_Assert(inputs[i]->size[2] == input.size[2] && inputs[i]->size[3] == input.size[3]);
CV_Assert(inputs[i].type() == input.type());
CV_Assert(inputs[i].dims == 4 && inputs[i].size[1] == input.size[1]);
CV_Assert(inputs[i].size[2] == input.size[2] && inputs[i].size[3] == input.size[3]);
}
Size outSize = Size(outputs[0].size[3], outputs[0].size[2]);
@@ -225,6 +200,14 @@ public:
return shape(out.area(), ksize);
}
virtual bool supportBackend(int backendId) CV_OVERRIDE
{
if (backendId == DNN_BACKEND_INFERENCE_ENGINE)
return preferableTarget != DNN_TARGET_MYRIAD || dilation.width == dilation.height;
else
return backendId == DNN_BACKEND_OPENCV || backendId == DNN_BACKEND_HALIDE;
}
bool getMemoryShapes(const std::vector<MatShape> &inputs,
const int requiredOutputs,
std::vector<MatShape> &outputs,
@@ -262,9 +245,9 @@ public:
return false;
}
virtual void finalize(const std::vector<Mat*> &inputs, std::vector<Mat> &outputs) CV_OVERRIDE
virtual void finalize(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr) CV_OVERRIDE
{
BaseConvolutionLayerImpl::finalize(inputs, outputs);
BaseConvolutionLayerImpl::finalize(inputs_arr, outputs_arr);
CV_Assert(!blobs.empty());
const int outCn = blobs[0].size[0];
@@ -1007,22 +990,24 @@ public:
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
Layer::forward_fallback(inputs_arr, outputs_arr, internals_arr);
}
if (inputs_arr.depth() == CV_16S)
{
forward_fallback(inputs_arr, outputs_arr, internals_arr);
return;
}
void forward(std::vector<Mat*> &inputs, std::vector<Mat> &outputs, std::vector<Mat> &internals) CV_OVERRIDE
{
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
/*printf("conv %s: input (%d x %d x %d x %d), kernel (%d x %d), pad (%d x %d), stride (%d x %d), dilation (%d x %d)\n",
name.c_str(), inputs[0]->size[0], inputs[0]->size[1], inputs[0]->size[2], inputs[0]->size[3],
name.c_str(), inputs[0].size[0], inputs[0].size[1], inputs[0].size[2], inputs[0].size[3],
kernel.width, kernel.height, pad.width, pad.height,
stride.width, stride.height, dilation.width, dilation.height);*/
CV_Assert_N(inputs.size() == (size_t)1, inputs[0]->size[1] % blobs[0].size[1] == 0,
outputs.size() == 1, inputs[0]->data != outputs[0].data);
CV_Assert_N(inputs.size() == (size_t)1, inputs[0].size[1] % blobs[0].size[1] == 0,
outputs.size() == 1, inputs[0].data != outputs[0].data);
int ngroups = inputs[0]->size[1]/blobs[0].size[1];
int ngroups = inputs[0].size[1]/blobs[0].size[1];
CV_Assert(outputs[0].size[1] % ngroups == 0);
int outCn = blobs[0].size[0];
@@ -1049,7 +1034,7 @@ public:
int nstripes = std::max(getNumThreads(), 1);
ParallelConv::run(*inputs[0], outputs[0], weightsMat, biasvec, reluslope,
ParallelConv::run(inputs[0], outputs[0], weightsMat, biasvec, reluslope,
kernel, pad, stride, dilation, activ.get(), ngroups, nstripes);
}
@@ -1089,6 +1074,29 @@ public:
return shape(ksize, inpH * inpW);
}
virtual bool supportBackend(int backendId) CV_OVERRIDE
{
#ifdef HAVE_INF_ENGINE
if (backendId == DNN_BACKEND_INFERENCE_ENGINE)
{
const int outGroupCn = blobs[0].size[1]; // Weights are in IOHW layout
const int group = numOutput / outGroupCn;
if (group != 1)
{
#if INF_ENGINE_VER_MAJOR_GE(INF_ENGINE_RELEASE_2018R3)
return preferableTarget == DNN_TARGET_CPU;
#endif
return false;
}
if (preferableTarget == DNN_TARGET_OPENCL || preferableTarget == DNN_TARGET_OPENCL_FP16)
return dilation.width == 1 && dilation.height == 1;
return true;
}
else
#endif // HAVE_INF_ENGINE
return backendId == DNN_BACKEND_OPENCV || backendId == DNN_BACKEND_HALIDE;
}
bool getMemoryShapes(const std::vector<MatShape> &inputs,
const int requiredOutputs,
std::vector<MatShape> &outputs,
@@ -1141,11 +1149,15 @@ public:
return false;
}
void finalize(const std::vector<Mat*> &inputs, std::vector<Mat> &outputs) CV_OVERRIDE
void finalize(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr) CV_OVERRIDE
{
BaseConvolutionLayerImpl::finalize(inputs, outputs);
BaseConvolutionLayerImpl::finalize(inputs_arr, outputs_arr);
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
getConvPoolPaddings(Size(outputs[0].size[3], outputs[0].size[2]),
Size(inputs[0]->size[3], inputs[0]->size[2]),
Size(inputs[0].size[3], inputs[0].size[2]),
kernel, stride, padMode, dilation, pad);
}
@@ -1494,18 +1506,21 @@ public:
CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget) &&
OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()),
forward_ocl(inputs_arr, outputs_arr, internals_arr))
forward_ocl(inputs_arr, outputs_arr, internals_arr));
Layer::forward_fallback(inputs_arr, outputs_arr, internals_arr);
}
if (inputs_arr.depth() == CV_16S)
{
forward_fallback(inputs_arr, outputs_arr, internals_arr);
return;
}
void forward(std::vector<Mat *> &inputs, std::vector<Mat> &outputs, std::vector<Mat> &internals) CV_OVERRIDE
{
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
std::vector<Mat> inputs, outputs, internals;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
internals_arr.getMatVector(internals);
int outCn = numOutput;
int inpCn = inputs[0]->size[1];
int inpCn = inputs[0].size[1];
bool is1x1flag = is1x1();
int nstripes = getNumThreads();
@@ -1520,13 +1535,13 @@ public:
int ngroups = outCn / blobs[0].size[1];
int inpGroupCn = inpCn / ngroups;
int outGroupCn = blobs[0].size[1];
const Mat& inp = *inputs[ii];
const Mat& inp = inputs[ii];
Mat& out = outputs[ii];
int numImg = inp.size[0];
int inpH = inp.size[2], inpW = inp.size[3];
int outH = out.size[2], outW = out.size[3];
Mat convBlob = inputs[ii]->reshape(1, numImg*inpCn);
Mat convBlob = inputs[ii].reshape(1, numImg*inpCn);
Mat decnBlob = out.reshape(1, numImg*outCn);
for (int n = 0; n < numImg; n++)