add alpha parameter to ELU

pull/21161/head
Smirnov Egor 3 years ago
parent d58b5ef74b
commit 4995aecd62
  1. 2
      modules/dnn/include/opencv2/dnn/all_layers.hpp
  2. 8
      modules/dnn/src/cuda/activations.cu
  3. 12
      modules/dnn/src/cuda/functors.hpp
  4. 2
      modules/dnn/src/cuda4dnn/kernels/activations.hpp
  5. 5
      modules/dnn/src/cuda4dnn/primitives/activation.hpp
  6. 20
      modules/dnn/src/layers/elementwise_layers.cpp
  7. 5
      modules/dnn/src/opencl/activations.cl

@ -545,6 +545,8 @@ CV__DNN_INLINE_NS_BEGIN
class CV_EXPORTS ELULayer : public ActivationLayer
{
public:
float alpha;
static Ptr<ELULayer> create(const LayerParams &params);
};

@ -119,8 +119,8 @@ void sigmoid(const Stream& stream, Span<T> output, View<T> input) {
}
template <class T>
void elu(const Stream& stream, Span<T> output, View<T> input) {
generic_op<T, ELUFunctor<T>>(stream, output, input);
void elu(const Stream& stream, Span<T> output, View<T> input, T alpha) {
generic_op<T, ELUFunctor<T>>(stream, output, input, {alpha});
}
template <class T>
@ -187,7 +187,7 @@ template void tanh<__half>(const Stream&, Span<__half>, View<__half>);
template void swish<__half>(const Stream&, Span<__half>, View<__half>);
template void mish<__half>(const Stream&, Span<__half>, View<__half>);
template void sigmoid<__half>(const Stream&, Span<__half>, View<__half>);
template void elu<__half>(const Stream&, Span<__half>, View<__half>);
template void elu<__half>(const Stream&, Span<__half>, View<__half>, __half);
template void abs<__half>(const Stream& stream, Span<__half> output, View<__half> input);
template void bnll<__half>(const Stream&, Span<__half>, View<__half>);
template void ceil<__half>(const Stream&, Span<__half>, View<__half>);
@ -207,7 +207,7 @@ template void tanh<float>(const Stream&, Span<float>, View<float>);
template void swish<float>(const Stream&, Span<float>, View<float>);
template void mish<float>(const Stream&, Span<float>, View<float>);
template void sigmoid<float>(const Stream&, Span<float>, View<float>);
template void elu<float>(const Stream&, Span<float>, View<float>);
template void elu<float>(const Stream&, Span<float>, View<float>, float);
template void abs<float>(const Stream& stream, Span<float> output, View<float> input);
template void bnll<float>(const Stream&, Span<float>, View<float>);
template void ceil<float>(const Stream&, Span<float>, View<float>);

@ -169,16 +169,20 @@ struct SigmoidFunctor {
template <class T>
struct ELUFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
CUDA4DNN_HOST_DEVICE Params() : alpha(1) { }
CUDA4DNN_HOST_DEVICE Params(T alpha_) : alpha(alpha_) { }
T alpha;
};
CUDA4DNN_DEVICE ELUFunctor() { }
CUDA4DNN_DEVICE ELUFunctor(const Params& params) { }
CUDA4DNN_DEVICE ELUFunctor() : ELUFunctor(Params{}) { }
CUDA4DNN_DEVICE ELUFunctor(const Params& params) : alpha{params.alpha} { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::expm1;
return value >= T(0) ? value : expm1(value);
return value >= T(0) ? value : alpha * expm1(value);
}
T alpha;
};
template <class T>

@ -34,7 +34,7 @@ namespace cv { namespace dnn { namespace cuda4dnn { namespace kernels {
void sigmoid(const csl::Stream& stream, csl::Span<T> output, csl::View<T> input);
template <class T>
void elu(const csl::Stream& stream, csl::Span<T> output, csl::View<T> input);
void elu(const csl::Stream& stream, csl::Span<T> output, csl::View<T> input, T alpha);
template <class T>
void abs(const csl::Stream& stream, csl::Span<T> output, csl::View<T> input);

@ -156,15 +156,16 @@ namespace cv { namespace dnn { namespace cuda4dnn {
template <class T>
class ELUOp final : public BaseOp<ELUOp, T> {
public:
ELUOp(csl::Stream stream_) : stream(std::move(stream_)) { }
ELUOp(csl::Stream stream_, T alpha_) : stream(std::move(stream_)), alpha(alpha_) { }
void calculate(csl::TensorSpan<T> output, csl::TensorView<T> input) const
{
kernels::elu<T>(stream, output, input);
kernels::elu<T>(stream, output, input, alpha);
}
private:
csl::Stream stream;
T alpha;
};
template <class T>

@ -987,6 +987,9 @@ const char* const SigmoidFunctor::BaseDefaultFunctor<SigmoidFunctor>::ocl_kernel
struct ELUFunctor : public BaseDefaultFunctor<ELUFunctor>
{
typedef ELULayer Layer;
float alpha;
explicit ELUFunctor(float alpha_ = 1.f) : alpha(alpha_) {}
bool supportBackend(int backendId, int)
{
@ -998,13 +1001,18 @@ struct ELUFunctor : public BaseDefaultFunctor<ELUFunctor>
inline float calculate(float x) const
{
return x >= 0.f ? x : exp(x) - 1.f;
return x >= 0.f ? x : alpha * (exp(x) - 1.f);
}
inline void setKernelParams(ocl::Kernel& kernel) const
{
kernel.set(3, alpha);
}
#ifdef HAVE_CUDA
Ptr<BackendNode> initCUDA(int target, csl::Stream stream)
{
return make_cuda_node<cuda4dnn::ELUOp>(target, stream);
return make_cuda_node<cuda4dnn::ELUOp>(target, stream, alpha);
}
#endif
@ -1012,7 +1020,7 @@ struct ELUFunctor : public BaseDefaultFunctor<ELUFunctor>
void attachHalide(const Halide::Expr& input, Halide::Func& top)
{
Halide::Var x("x"), y("y"), c("c"), n("n");
top(x, y, c, n) = select(input >= 0.0f, input, exp(input) - 1);
top(x, y, c, n) = select(input >= 0.0f, input, alpha * (exp(input) - 1));
}
#endif // HAVE_HALIDE
@ -1026,7 +1034,7 @@ struct ELUFunctor : public BaseDefaultFunctor<ELUFunctor>
#ifdef HAVE_DNN_NGRAPH
std::shared_ptr<ngraph::Node> initNgraphAPI(const std::shared_ptr<ngraph::Node>& node)
{
return std::make_shared<ngraph::op::Elu>(node, 1.0);
return std::make_shared<ngraph::op::Elu>(node, alpha);
}
#endif // HAVE_DNN_NGRAPH
@ -1856,8 +1864,10 @@ Ptr<SigmoidLayer> SigmoidLayer::create(const LayerParams& params)
Ptr<ELULayer> ELULayer::create(const LayerParams& params)
{
Ptr<ELULayer> l(new ElementWiseLayer<ELUFunctor>(ELUFunctor()));
float alpha = params.get<float>("alpha", 1.0f);
Ptr<ELULayer> l(new ElementWiseLayer<ELUFunctor>(ELUFunctor(alpha)));
l->setParamsFrom(params);
l->alpha = alpha;
return l;
}

@ -131,13 +131,14 @@ __kernel void PowForward(const int n, __global const T* in, __global T* out,
out[index] = pow(shift + scale * in[index], power);
}
__kernel void ELUForward(const int n, __global const T* in, __global T* out)
__kernel void ELUForward(const int n, __global const T* in, __global T* out,
const KERNEL_ARG_DTYPE alpha)
{
int index = get_global_id(0);
if (index < n)
{
T src = in[index];
out[index] = (src >= 0.f) ? src : exp(src) - 1;
out[index] = (src >= 0.f) ? src : alpha * (exp(src) - 1);
}
}

Loading…
Cancel
Save