IntenseWebs/openvino
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[GPU] ROIAlign v9 support (#11899)

Browse Source 
* ROIAlign v9 support

* Code changes after review1

* Code changes after review2

* fix of single layer test for Windows

* Since PR #12043 we don't need strong include order of primitive_base.hpp and
impls/implementation map.hpp anymore

* Code changes after review3

* Code changes after review4
This commit is contained in:
Konstantin Beluchenko 2022-07-21 12:52:32 +03:00 committed by GitHub
parent 598975fb63
commit ed65d199bb
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
15 changed files with 639 additions and 169 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
src/plugins/intel_gpu/include/intel_gpu/plugin/primitives_list.hpp
View File

@ -235,6 +235,7 @@ REGISTER_FACTORY(v8, AdaptiveMaxPool);
// ------------------------------ Supported v9 ops ------------------------------ //
REGISTER_FACTORY(v9, SoftSign)
REGISTER_FACTORY(v9, ROIAlign);
// --------------------------- Supported internal ops --------------------------- //
REGISTER_FACTORY(internal, NonMaxSuppressionIEInternal);

31
src/plugins/intel_gpu/include/intel_gpu/primitives/roi_align.hpp
View File

@ -19,10 +19,10 @@ struct roi_align : public primitive_base<roi_align> {
CLDNN_DECLARE_PRIMITIVE(roi_align)
/// @brief Pooling mode for the @ref roi_align
enum PoolingMode {
Max,
Avg
};
enum PoolingMode { max, avg };
/// @brief Aligned mode for the @ref roi_align
enum AlignedMode { asymmetric, half_pixel_for_nn, half_pixel };
/// @brief Constructs roi_align primitive.
/// @param id This primitive id.
@ -32,24 +32,25 @@ struct roi_align : public primitive_base<roi_align> {
/// @param sampling_ratio Number of bins over height and width to use to calculate each output feature map element.
/// @param spatial_scale multiplicative spatial scale factor to translate ROI coordinates
/// from their input spatial scale to the scale used when pooling.
/// @param mode Method to perform pooling to produce output feature map elements.
/// @param shrink_axis_mask Array of bits, that provide shrinks the dimensionality by 1, taking on the value at index begin[i].
/// @param pooling_mode Method to perform pooling to produce output feature map elements.
/// @param aligned_mode Method to coordinates alignment.
roi_align(const primitive_id& id,
const std::vector<primitive_id>& inputs,
int pooled_h,
int pooled_w,
int sampling_ratio,
float spatial_scale,
PoolingMode mode,
PoolingMode pooling_mode,
AlignedMode aligned_mode,
const primitive_id& ext_prim_id = "",
const padding& output_padding = padding())
: primitive_base(id, inputs, ext_prim_id, output_padding),
pooled_h {pooled_h},
pooled_w {pooled_w},
sampling_ratio {sampling_ratio},
spatial_scale {spatial_scale},
mode {mode}
{}
pooled_h{pooled_h},
pooled_w{pooled_w},
sampling_ratio{sampling_ratio},
spatial_scale{spatial_scale},
pooling_mode{pooling_mode},
aligned_mode{aligned_mode} {}
/// @brief Height of the ROI output feature map.
int pooled_h;
@ -61,7 +62,9 @@ struct roi_align : public primitive_base<roi_align> {
/// from their input spatial scale to the scale used when pooling.
float spatial_scale;
/// @brief Method to perform pooling to produce output feature map elements.
PoolingMode mode;
PoolingMode pooling_mode;
/// @brief Method to coordinate alignment.
AlignedMode aligned_mode;
};
/// @}
/// @}

16
src/plugins/intel_gpu/src/graph/impls/implementation_map.hpp
View File

@ -204,11 +204,27 @@ public:
return false;
}
static void add(impl_types impl_type, factory_type factory,
const std::vector<data_types>& types, const std::vector<format::type>& formats) {
add(impl_type, factory, combine(types, formats));
}
static void add(impl_types impl_type, factory_type factory, std::set<key_type> keys) {
if (impl_type == impl_types::any) {
throw std::runtime_error("[CLDNN] Can't register impl with type any");
}
map_type::instance().insert({impl_type, {keys, factory}});
}
private:
static std::set<key_type> combine(const std::vector<data_types>& types, const std::vector<format::type>& formats) {
std::set<key_type> keys;
for (const auto& type : types) {
for (const auto& format : formats) {
keys.emplace(type, format);
}
}
return keys;
}
};
} // namespace cldnn

46
src/plugins/intel_gpu/src/graph/impls/ocl/roi_align.cpp
View File

@ -1,13 +1,14 @@
// Copyright (C) 2018-2022 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
//
#include "roi_align_inst.h"
#include "primitive_base.hpp"
#include "impls/implementation_map.hpp"
#include "intel_gpu/runtime/error_handler.hpp"
#include "kernel_selector_helper.h"
#include "roi_align/roi_align_kernel_selector.h"
#include "roi_align/roi_align_kernel_ref.h"
#include "roi_align/roi_align_kernel_selector.h"
#include "roi_align_inst.h"
namespace cldnn {
namespace ocl {
@ -15,13 +16,25 @@ namespace ocl {
namespace {
kernel_selector::pool_type from(roi_align::PoolingMode mode) {
switch (mode) {
case roi_align::PoolingMode::Max:
case roi_align::PoolingMode::max:
return kernel_selector::pool_type::MAX;
default:
case roi_align::PoolingMode::Avg:
case roi_align::PoolingMode::avg:
return kernel_selector::pool_type::AVG;
}
}
kernel_selector::roi_aligned_mode from(roi_align::AlignedMode mode) {
switch (mode) {
case roi_align::AlignedMode::half_pixel_for_nn:
return kernel_selector::roi_aligned_mode::HALF_PIXEL_FOR_NN;
case roi_align::AlignedMode::half_pixel:
return kernel_selector::roi_aligned_mode::HALF_PIXEL;
default:
case roi_align::AlignedMode::asymmetric:
return kernel_selector::roi_aligned_mode::ASYMMETRIC;
}
}
} // namespace
struct roi_align_impl : typed_primitive_impl_ocl<roi_align> {
@ -35,8 +48,8 @@ struct roi_align_impl : typed_primitive_impl_ocl<roi_align> {
protected:
kernel_arguments_data get_arguments(typed_primitive_inst<roi_align>& instance, int32_t) const override {
kernel_arguments_data args;
args.inputs = { instance.input_memory_ptr(), instance.rois_memory(), instance.batches_memory() };
args.outputs = { instance.output_memory_ptr() };
args.inputs = {instance.input_memory_ptr(), instance.rois_memory(), instance.batches_memory()};
args.outputs = {instance.output_memory_ptr()};
return args;
}
@ -67,10 +80,10 @@ public:
auto roi_align_optional_params =
get_default_optional_params<kernel_selector::roi_align_optional_params>(arg.get_program());
const auto roi_bfyx = convert_data_tensor(rois_layout);
roi_align_params.inputs.push_back(roi_bfyx.FlattenFeatureAndSpatials());
roi_align_params.inputs.push_back(convert_data_tensor(rois_layout));
roi_align_params.inputs.push_back(convert_data_tensor(batches_layout));
roi_align_params.mode = from(primitive->mode);
roi_align_params.pooling_mode = from(primitive->pooling_mode);
roi_align_params.aligned_mode = from(primitive->aligned_mode);
roi_align_params.sampling_ratio = primitive->sampling_ratio;
roi_align_params.spatial_scale = primitive->spatial_scale;
@ -91,11 +104,16 @@ public:
namespace detail {
attach_roi_align_impl::attach_roi_align_impl() {
implementation_map<roi_align>::add(impl_types::ocl, roi_align_impl::create,
{
std::make_tuple(data_types::f16, format::bfyx),
std::make_tuple(data_types::f32, format::bfyx),
});
auto types = {data_types::f16, data_types::f32, data_types::i8, data_types::u8, data_types::i32};
auto formats = {format::bfyx,
format::b_fs_yx_fsv16,
format::b_fs_yx_fsv32,
format::bs_fs_yx_bsv16_fsv16,
format::bs_fs_yx_bsv32_fsv16,
format::bs_fs_yx_bsv32_fsv32};
implementation_map<roi_align>::add(impl_types::ocl, roi_align_impl::create, types, formats);
}
} // namespace detail

7
src/plugins/intel_gpu/src/graph/program.cpp
View File

@ -1400,7 +1400,8 @@ void program::set_layout_optimizer_attributes(layout_optimizer& lo) {
prim.type() != cldnn::mvn::type_id() &&
prim.type() != cldnn::gather::type_id() &&
prim.type() != cldnn::scatter_nd_update::type_id() &&
prim.type() != cldnn::non_max_suppression::type_id()) {
prim.type() != cldnn::non_max_suppression::type_id() &&
prim.type() != cldnn::roi_align::type_id()) {
can_use_fsv16 = false;
}
@ -1428,9 +1429,11 @@ void program::set_layout_optimizer_attributes(layout_optimizer& lo) {
prim.type() != cldnn::generic_layer::type_id() &&
prim.type() != cldnn::scatter_nd_update::type_id() &&
prim.type() != cldnn::quantize::type_id() &&
prim.type() != cldnn::non_max_suppression::type_id())
prim.type() != cldnn::non_max_suppression::type_id() &&
prim.type() != cldnn::roi_align::type_id()) {
can_use_bs_fs_yx_bsv16_fsv16 = false;
}
}
size_t total_conv_layers = lo.get_total_conv_count();
// Due to fact that single winograd convolution is faster than b_fs_yx_fsv16 and

25
src/plugins/intel_gpu/src/graph/roi_align.cpp
View File

@ -6,6 +6,7 @@
#include "primitive_type_base.h"
#include <sstream>
#include <json_object.h>
#include "openvino/core/enum_names.hpp"
namespace cldnn {
@ -23,7 +24,9 @@ layout roi_align_inst::calc_output_layout(roi_align_node const& node) {
auto rois_layout = node.input(1).get_output_layout();
auto num_rois = rois_layout.batch();
auto num_channels = input_layout.feature();
return layout(input_layout.data_type, format::bfyx, {num_rois, num_channels, primitive->pooled_h, primitive->pooled_w});
return layout(input_layout.data_type,
input_layout.format,
{num_rois, num_channels, primitive->pooled_h, primitive->pooled_w});
}
std::string roi_align_inst::to_string(roi_align_node const& node) {
@ -36,7 +39,8 @@ std::string roi_align_inst::to_string(roi_align_node const& node) {
roi_align_info.add("pooled_w", node.get_primitive()->pooled_w);
roi_align_info.add("sampling_ratio", node.get_primitive()->sampling_ratio);
roi_align_info.add("spatial_scale", node.get_primitive()->spatial_scale);
roi_align_info.add("mode", node.get_primitive()->mode == roi_align::PoolingMode::Max ? "Max" : "Avg");
roi_align_info.add("pooling_mode", ov::as_string(node.get_primitive()->pooling_mode));
roi_align_info.add("aligned_mode", ov::as_string(node.get_primitive()->aligned_mode));
node_info->add("roi_align info", roi_align_info);
std::stringstream primitive_description;
node_info->dump(primitive_description);
@ -44,3 +48,20 @@ std::string roi_align_inst::to_string(roi_align_node const& node) {
}
} // namespace cldnn
namespace ov {
template <> EnumNames<roi_align::PoolingMode>& EnumNames<roi_align::PoolingMode>::get() {
static auto enum_names =
EnumNames<roi_align::PoolingMode>("PoolingMode", {{"max", roi_align::PoolingMode::max},
{"avg", roi_align::PoolingMode::avg}});
return enum_names;
}
template <> EnumNames<roi_align::AlignedMode>& EnumNames<roi_align::AlignedMode>::get() {
static auto enum_names =
EnumNames<roi_align::AlignedMode>("AlignedMode", {{"asymmetric", roi_align::AlignedMode::asymmetric},
{"half_pixel_for_nn", roi_align::AlignedMode::half_pixel_for_nn},
{"half_pixel", roi_align::AlignedMode::half_pixel}});
return enum_names;
}
} // namespace ov

33
src/plugins/intel_gpu/src/kernel_selector/core/actual_kernels/roi_align/roi_align_kernel_ref.cpp
View File

@ -10,11 +10,13 @@ ParamsKey ROIAlignKernelRef::GetSupportedKey() const {
ParamsKey k;
k.EnableInputDataType(Datatype::F16);
k.EnableInputDataType(Datatype::F32);
k.EnableInputDataType(Datatype::UINT8);
k.EnableInputDataType(Datatype::INT8);
k.EnableInputDataType(Datatype::INT32);
k.EnableOutputDataType(Datatype::F16);
k.EnableOutputDataType(Datatype::F32);
k.EnableInputLayout(DataLayout::bfyx);
k.EnableOutputLayout(DataLayout::bfyx);
k.EnableAllInputLayout();
k.EnableAllOutputLayout();
k.EnableTensorOffset();
k.EnableTensorPitches();
k.EnableBatching();
@ -70,20 +72,35 @@ bool ROIAlignKernelRef::Validate(const Params& p, const optional_params& o) cons
if (params.inputs.size() != 3)
return false;
if (params.outputs[0].Dimentions() > 4 || params.inputs[0].Dimentions() > 4 || params.inputs[1].Dimentions() > 2)
return false;
return true;
}
JitConstants ROIAlignKernelRef::GetJitConstants(const roi_align_params &params) const {
JitConstants ROIAlignKernelRef::GetJitConstants(const roi_align_params& params) const {
JitConstants jit = MakeBaseParamsJitConstants(params);
jit.AddConstant(MakeJitConstant("SPATIAL_SCALE", params.spatial_scale));
jit.AddConstant(MakeJitConstant("SAMPLING_RATIO", params.sampling_ratio));
if (params.mode == PoolType::MAX)
if (params.pooling_mode == PoolType::MAX) {
jit.AddConstant(MakeJitConstant("MAX_POOL", true));
else if (params.mode == PoolType::AVG)
} else if (params.pooling_mode == PoolType::AVG) {
jit.AddConstant(MakeJitConstant("AVG_POOL", true));
}
if (params.aligned_mode == roi_aligned_mode::ASYMMETRIC) {
jit.AddConstant(MakeJitConstant("OFFSET_SRC", 0.f));
jit.AddConstant(MakeJitConstant("OFFSET_DST", 0.f));
jit.AddConstant(MakeJitConstant("MIN_SIZE", 1.0f));
} else if (params.aligned_mode == roi_aligned_mode::HALF_PIXEL_FOR_NN) {
jit.AddConstant(MakeJitConstant("OFFSET_SRC", 0.f));
jit.AddConstant(MakeJitConstant("OFFSET_DST", -0.5f));
jit.AddConstant(MakeJitConstant("MIN_SIZE", 0.f));
} else if (params.aligned_mode == roi_aligned_mode::HALF_PIXEL) {
jit.AddConstant(MakeJitConstant("OFFSET_SRC", 0.5f));
jit.AddConstant(MakeJitConstant("OFFSET_DST", -0.5f));
jit.AddConstant(MakeJitConstant("MIN_SIZE", 0.f));
}
return jit;
}

7
src/plugins/intel_gpu/src/kernel_selector/core/actual_kernels/roi_align/roi_align_kernel_ref.h
View File

@ -7,16 +7,19 @@
namespace kernel_selector {
enum class roi_aligned_mode : uint32_t { ASYMMETRIC, HALF_PIXEL_FOR_NN, HALF_PIXEL };
struct roi_align_params : public base_params {
roi_align_params() : base_params{KernelType::ROI_ALIGN} {}
int sampling_ratio = 0;
float spatial_scale = 1.f;
PoolType mode = PoolType::MAX;
PoolType pooling_mode = PoolType::MAX;
roi_aligned_mode aligned_mode = roi_aligned_mode::ASYMMETRIC;
ParamsKey GetParamsKey() const override {
auto k = base_params::GetParamsKey();
k.EnablePoolType(mode);
k.EnablePoolType(pooling_mode);
return k;
}
};

100
src/plugins/intel_gpu/src/kernel_selector/core/cl_kernels/roi_align_ref.cl
View File

@ -5,20 +5,17 @@
#include "include/batch_headers/common.cl"
#include "include/batch_headers/data_types.cl"
#define MAX(a,b) ((a) > (b) ? (a) : (b))
#define MAX(a, b) ((a) > (b) ? (a) : (b))
#define NUM_ROIS OUTPUT_BATCH_NUM
#define NUM_CHANNELS INPUT0_FEATURE_NUM
#define POOLED_WIDTH OUTPUT_SIZE_X
#define POOLED_HEIGHT OUTPUT_SIZE_Y
KERNEL(roi_align_ref)
(
const __global INPUT0_TYPE * src_data,
__global OUTPUT_TYPE * dst_data,
const __global INPUT1_TYPE * src_rois,
const __global INPUT2_TYPE * src_batches
)
{
(const __global INPUT0_TYPE* src_data,
__global OUTPUT_TYPE* dst_data,
const __global INPUT1_TYPE* src_rois,
const __global INPUT2_TYPE* src_batches) {
const size_t i = get_global_id(0);
const uint x = i % POOLED_WIDTH;
@ -26,78 +23,85 @@ KERNEL(roi_align_ref)
const uint c = i / POOLED_WIDTH / POOLED_HEIGHT % NUM_CHANNELS;
const uint r = i / POOLED_WIDTH / POOLED_HEIGHT / NUM_CHANNELS % NUM_ROIS;
const __global INPUT1_TYPE* roi_ptr = &src_rois[INPUT1_BATCH_PITCH * r];
const __global INPUT1_TYPE* roi_ptr = &src_rois[INPUT1_GET_INDEX(r, 0, 0, 0)];
// Get ROI`s corners
const INPUT1_TYPE x1 = *roi_ptr * (INPUT1_TYPE) SPATIAL_SCALE;
const INPUT1_TYPE y1 = roi_ptr[1] * (INPUT1_TYPE) SPATIAL_SCALE;
const INPUT1_TYPE x2 = roi_ptr[2] * (INPUT1_TYPE) SPATIAL_SCALE;
const INPUT1_TYPE y2 = roi_ptr[3] * (INPUT1_TYPE) SPATIAL_SCALE;
const INPUT1_TYPE x1 =
(roi_ptr[0] + (INPUT1_TYPE)OFFSET_SRC) * (INPUT1_TYPE)SPATIAL_SCALE + (INPUT1_TYPE)OFFSET_DST;
const INPUT1_TYPE y1 =
(roi_ptr[1] + (INPUT1_TYPE)OFFSET_SRC) * (INPUT1_TYPE)SPATIAL_SCALE + (INPUT1_TYPE)OFFSET_DST;
const INPUT1_TYPE x2 =
(roi_ptr[2] + (INPUT1_TYPE)OFFSET_SRC) * (INPUT1_TYPE)SPATIAL_SCALE + (INPUT1_TYPE)OFFSET_DST;
const INPUT1_TYPE y2 =
(roi_ptr[3] + (INPUT1_TYPE)OFFSET_SRC) * (INPUT1_TYPE)SPATIAL_SCALE + (INPUT1_TYPE)OFFSET_DST;
const INPUT1_TYPE roi_width = MAX(x2 - x1, (INPUT1_TYPE) 1.0);
const INPUT1_TYPE roi_height = MAX(y2 - y1, (INPUT1_TYPE) 1.0);
const INPUT1_TYPE roi_width = MAX(x2 - x1, (INPUT1_TYPE)MIN_SIZE);
const INPUT1_TYPE roi_height = MAX(y2 - y1, (INPUT1_TYPE)MIN_SIZE);
const INPUT1_TYPE bin_width = roi_width / POOLED_WIDTH;
const INPUT1_TYPE bin_height = roi_height / POOLED_HEIGHT;
const int sampling_ratio_x = SAMPLING_RATIO == 0 ? (int) ceil(bin_width) : SAMPLING_RATIO;
const int sampling_ratio_y = SAMPLING_RATIO == 0 ? (int) ceil(bin_height) : SAMPLING_RATIO;
const int sampling_ratio_x = SAMPLING_RATIO == 0 ? (int)ceil(bin_width) : SAMPLING_RATIO;
const int sampling_ratio_y = SAMPLING_RATIO == 0 ? (int)ceil(bin_height) : SAMPLING_RATIO;
const INPUT1_TYPE sample_distance_x = bin_width / (INPUT1_TYPE) sampling_ratio_x;
const INPUT1_TYPE sample_distance_y = bin_height / (INPUT1_TYPE) sampling_ratio_y;
const INPUT1_TYPE sample_distance_x = bin_width / (INPUT1_TYPE)sampling_ratio_x;
const INPUT1_TYPE sample_distance_y = bin_height / (INPUT1_TYPE)sampling_ratio_y;
const __global INPUT0_TYPE* data = src_data + INPUT0_GET_INDEX(r, c, 0, 0);
const __global INPUT0_TYPE* data = src_data + INPUT0_OFFSET + r*INPUT0_BATCH_PITCH + INPUT0_FEATURE_PITCH*c;
OUTPUT_TYPE pooled_value = 0;
for (unsigned int y_sample_ind = 0; y_sample_ind < sampling_ratio_y; y_sample_ind++) {
INPUT1_TYPE sample_y = y1 + (INPUT1_TYPE) y * bin_height +
sample_distance_y * ((INPUT1_TYPE) y_sample_ind + (INPUT1_TYPE) 0.5f);
INPUT1_TYPE sample_y =
y1 + (INPUT1_TYPE)y * bin_height + sample_distance_y * ((INPUT1_TYPE)y_sample_ind + (INPUT1_TYPE)0.5f);
for (unsigned int x_sample_ind = 0; x_sample_ind < sampling_ratio_x; x_sample_ind++) {
INPUT1_TYPE sample_x = x1 + (INPUT1_TYPE) x * bin_width +
sample_distance_x * ((INPUT1_TYPE) x_sample_ind + (INPUT1_TYPE) 0.5f);
INPUT1_TYPE sample_x =
x1 + (INPUT1_TYPE)x * bin_width + sample_distance_x * ((INPUT1_TYPE)x_sample_ind + (INPUT1_TYPE)0.5f);
unsigned int sample_y_low = 0;
unsigned int sample_x_low = 0;
unsigned int sample_y_high = 0;
unsigned int sample_x_high = 0;
INPUT1_TYPE weight_left = (INPUT1_TYPE) 0.f;
INPUT1_TYPE weight_right = (INPUT1_TYPE) 0.f;
INPUT1_TYPE weight_top = (INPUT1_TYPE) 0.f;
INPUT1_TYPE weight_bottom = (INPUT1_TYPE) 0.f;
INPUT1_TYPE weight_left = INPUT1_VAL_ZERO;
INPUT1_TYPE weight_right = INPUT1_VAL_ZERO;
INPUT1_TYPE weight_top = INPUT1_VAL_ZERO;
INPUT1_TYPE weight_bottom = INPUT1_VAL_ZERO;
if (sample_x >= -1.0 || sample_x <= INPUT0_SIZE_X || sample_y >= -1.0 || sample_y <= INPUT0_SIZE_Y) {
sample_x = MAX(sample_x, (INPUT1_TYPE) 0.f);
sample_y = MAX(sample_y, (INPUT1_TYPE) 0.f);
sample_x = MAX(sample_x, INPUT1_VAL_ZERO);
sample_y = MAX(sample_y, INPUT1_VAL_ZERO);
sample_y_low = (unsigned int) sample_y;
sample_x_low = (unsigned int) sample_x;
sample_y_low = (unsigned int)sample_y;
sample_x_low = (unsigned int)sample_x;
if (sample_y_low >= INPUT0_SIZE_Y - 1) {
sample_y_high = sample_y_low = INPUT0_SIZE_Y - 1;
sample_y = (INPUT1_TYPE) sample_y_low;
sample_y = (INPUT1_TYPE)sample_y_low;
} else {
sample_y_high = sample_y_low + 1;
}
if (sample_x_low >= INPUT0_SIZE_X - 1) {
sample_x_high = sample_x_low = INPUT0_SIZE_X - 1;
sample_x = (INPUT1_TYPE) sample_x_low;
sample_x = (INPUT1_TYPE)sample_x_low;
} else {
sample_x_high = sample_x_low + 1;
}
// weight calculation for bilinear interpolation
weight_top = sample_y - (INPUT1_TYPE) sample_y_low;
weight_left = sample_x - (INPUT1_TYPE) sample_x_low;
weight_bottom = (INPUT1_TYPE) 1.f - weight_top;
weight_right = (INPUT1_TYPE) 1.f - weight_left;
weight_top = sample_y - (INPUT1_TYPE)sample_y_low;
weight_left = sample_x - (INPUT1_TYPE)sample_x_low;
weight_bottom = INPUT1_VAL_ONE - weight_top;
weight_right = INPUT1_VAL_ONE - weight_left;
}
const INPUT0_TYPE top_left = data[sample_y_low * INPUT0_Y_PITCH + sample_x_low * INPUT0_X_PITCH];
const INPUT0_TYPE top_right = data[sample_y_low * INPUT0_Y_PITCH + sample_x_high * INPUT0_X_PITCH];
const INPUT0_TYPE bottom_left = data[sample_y_high * INPUT0_Y_PITCH + sample_x_low * INPUT0_X_PITCH];
const INPUT0_TYPE bottom_right = data[sample_y_high * INPUT0_Y_PITCH + sample_x_high * INPUT0_X_PITCH];
const INPUT0_TYPE interpolated_value = weight_bottom * weight_right * top_left +
weight_bottom * weight_left * top_right +
weight_top * weight_right * bottom_left +
weight_top * weight_left * bottom_right;
const INPUT0_TYPE top_left = data[INPUT0_GET_INDEX(0, 0, sample_y_low, sample_x_low)];
const INPUT0_TYPE top_right = data[INPUT0_GET_INDEX(0, 0, sample_y_low, sample_x_high)];
const INPUT0_TYPE bottom_left = data[INPUT0_GET_INDEX(0, 0, sample_y_high, sample_x_low)];
const INPUT0_TYPE bottom_right = data[INPUT0_GET_INDEX(0, 0, sample_y_high, sample_x_high)];
const INPUT0_TYPE interpolated_value =
weight_bottom * weight_right * top_left + weight_bottom * weight_left * top_right +
weight_top * weight_right * bottom_left + weight_top * weight_left * bottom_right;
#if MAX_POOL
pooled_value = MAX(pooled_value, interpolated_value);
#elif AVG_POOL
@ -108,7 +112,7 @@ KERNEL(roi_align_ref)
#if AVG_POOL
pooled_value /= sampling_ratio_x * sampling_ratio_x;
#endif
const uint output_offset = OUTPUT_OFFSET + x*OUTPUT_X_PITCH + y*OUTPUT_Y_PITCH + c*OUTPUT_FEATURE_PITCH + r*OUTPUT_BATCH_PITCH;
dst_data[output_offset] = ACTIVATION((OUTPUT_TYPE)pooled_value, ACTIVATION_PARAMS);
dst_data[OUTPUT_GET_INDEX(r, c, y, x)] = ACTIVATION((OUTPUT_TYPE)pooled_value, ACTIVATION_PARAMS);
}

50
src/plugins/intel_gpu/src/plugin/ops/roi_align.cpp
View File

@ -1,29 +1,44 @@
// Copyright (C) 2018-2022 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
//
#include "intel_gpu/plugin/program.hpp"
#include "intel_gpu/plugin/common_utils.hpp"
#include "ngraph/op/roi_align.hpp"
#include "intel_gpu/primitives/roi_align.hpp"
#include <memory>
#include "intel_gpu/plugin/common_utils.hpp"
#include "intel_gpu/plugin/program.hpp"
#include "intel_gpu/primitives/roi_align.hpp"
namespace ov {
namespace intel_gpu {
namespace {
cldnn::roi_align::PoolingMode from(ngraph::op::v3::ROIAlign::PoolingMode mode) {
template <typename T>
cldnn::roi_align::PoolingMode from(T mode) {
switch (mode) {
case ngraph::op::v3::ROIAlign::PoolingMode::MAX:
return cldnn::roi_align::PoolingMode::Max;
case ngraph::op::v3::ROIAlign::PoolingMode::AVG:
case T::MAX:
return cldnn::roi_align::PoolingMode::max;
case T::AVG:
default:
return cldnn::roi_align::PoolingMode::Avg;
return cldnn::roi_align::PoolingMode::avg;
}
}
cldnn::roi_align::AlignedMode from(ngraph::op::v9::ROIAlign::AlignedMode mode) {
switch (mode) {
case ngraph::op::v9::ROIAlign::AlignedMode::HALF_PIXEL_FOR_NN:
return cldnn::roi_align::AlignedMode::half_pixel_for_nn;
case ngraph::op::v9::ROIAlign::AlignedMode::HALF_PIXEL:
return cldnn::roi_align::AlignedMode::half_pixel;
case ngraph::op::v9::ROIAlign::AlignedMode::ASYMMETRIC:
default:
return cldnn::roi_align::AlignedMode::asymmetric;
}
}
void CreateROIAlignOp(Program& p, const std::shared_ptr<ngraph::op::v3::ROIAlign>& op) {
p.ValidateInputs(op, { 3 });
p.ValidateInputs(op, {3});
auto roi_align_prim = cldnn::roi_align(layer_type_name_ID(op),
p.GetInputPrimitiveIDs(op),
op->get_pooled_h(),
@ -31,6 +46,22 @@ void CreateROIAlignOp(Program& p, const std::shared_ptr<ngraph::op::v3::ROIAlign
op->get_sampling_ratio(),
op->get_spatial_scale(),
from(op->get_mode()),
cldnn::roi_align::AlignedMode::asymmetric,
op->get_friendly_name());
p.AddPrimitive(roi_align_prim);
p.AddPrimitiveToProfiler(op);
}
void CreateROIAlignOp(Program& p, const std::shared_ptr<ngraph::op::v9::ROIAlign>& op) {
p.ValidateInputs(op, {3});
auto roi_align_prim = cldnn::roi_align(layer_type_name_ID(op),
p.GetInputPrimitiveIDs(op),
op->get_pooled_h(),
op->get_pooled_w(),
op->get_sampling_ratio(),
op->get_spatial_scale(),
from(op->get_mode()),
from(op->get_aligned_mode()),
op->get_friendly_name());
p.AddPrimitive(roi_align_prim);
p.AddPrimitiveToProfiler(op);
@ -39,6 +70,7 @@ void CreateROIAlignOp(Program& p, const std::shared_ptr<ngraph::op::v3::ROIAlign
} // anonymous namespace
REGISTER_FACTORY_IMPL(v3, ROIAlign);
REGISTER_FACTORY_IMPL(v9, ROIAlign);
} // namespace intel_gpu
} // namespace ov

178
src/plugins/intel_gpu/tests/test_cases/roi_align_gpu_test.cpp Normal file
View File

@ -0,0 +1,178 @@
// Copyright (C) 2018-2022 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
//
///////////////////////////////////////////////////////////////////////////////////////////////////
#include <intel_gpu/primitives/data.hpp>
#include <intel_gpu/primitives/input_layout.hpp>
#include <intel_gpu/primitives/roi_align.hpp>
#include "test_utils.h"
using namespace cldnn;
using namespace ::tests;
template <typename TD, typename TI, cldnn::format::type fmt>
struct TypesWithFormat {
using DataType = TD;
using IndexType = TI;
static const cldnn::format::type format = fmt;
};
template <typename Types>
struct roi_align_test : public testing::Test {
using TD = typename Types::DataType;
using TI = typename Types::IndexType;
const data_types device_data_type = type_to_data_type<TD>::value;
const data_types device_ind_type = type_to_data_type<TI>::value;
const cldnn::format::type blocked_format = Types::format;
const cldnn::format::type plain_format = format::bfyx;
const int pooled_h{2};
const int pooled_w{2};
const int sampling_ratio{2};
const float spatial_scale{1};
const std::vector<TD> input_data = {
TD(0.f), TD(1.f), TD(8.f), TD(5.f), TD(5.f), TD(2.f), TD(0.f), TD(7.f), TD(7.f), TD(10.f), TD(4.f),
TD(5.f), TD(9.f), TD(0.f), TD(0.f), TD(5.f), TD(7.f), TD(0.f), TD(4.f), TD(0.f), TD(4.f), TD(7.f),
TD(6.f), TD(10.f), TD(9.f), TD(5.f), TD(1.f), TD(7.f), TD(4.f), TD(7.f), TD(10.f), TD(8.f), TD(2.f),
TD(0.f), TD(8.f), TD(3.f), TD(6.f), TD(8.f), TD(10.f), TD(4.f), TD(2.f), TD(10.f), TD(7.f), TD(8.f),
TD(7.f), TD(0.f), TD(6.f), TD(9.f), TD(2.f), TD(4.f), TD(8.f), TD(5.f), TD(2.f), TD(3.f), TD(3.f),
TD(1.f), TD(5.f), TD(9.f), TD(10.f), TD(0.f), TD(9.f), TD(5.f), TD(5.f), TD(3.f), TD(10.f), TD(5.f),
TD(2.f), TD(0.f), TD(10.f), TD(0.f), TD(5.f), TD(4.f), TD(3.f), TD(10.f), TD(5.f), TD(5.f), TD(10.f),
TD(0.f), TD(8.f), TD(8.f), TD(9.f), TD(1.f), TD(0.f), TD(7.f), TD(9.f), TD(6.f), TD(8.f), TD(7.f),
TD(10.f), TD(9.f), TD(2.f), TD(3.f), TD(3.f), TD(5.f), TD(6.f), TD(9.f), TD(4.f), TD(9.f), TD(2.f),
TD(4.f), TD(5.f), TD(5.f), TD(3.f), TD(1.f), TD(1.f), TD(6.f), TD(8.f), TD(0.f), TD(5.f), TD(5.f),
TD(10.f), TD(8.f), TD(6.f), TD(9.f), TD(6.f), TD(9.f), TD(1.f), TD(2.f), TD(7.f), TD(1.f), TD(1.f),
TD(3.f), TD(0.f), TD(4.f), TD(0.f), TD(7.f), TD(10.f), TD(2.f)};
const std::vector<TD> coords_data = {TD(2.f), TD(2.f), TD(4.f), TD(4.f), TD(2.f), TD(2.f), TD(4.f), TD(4.f)};
const std::vector<TI> roi_data = {0, 1};
const layout input_lt = layout(device_data_type, plain_format, {2, 1, 8, 8});
const layout coords_lt = layout(device_data_type, plain_format, {2, 4, 1, 1});
const layout roi_lt = layout(device_ind_type, plain_format, {2, 1, 1, 1});
memory::ptr get_memory(engine& engine, const layout& lt, const std::vector<TD>& data) const {
auto mem = engine.allocate_memory(lt);
tests::set_values(mem, data);
return mem;
}
memory::ptr get_roi_memory(engine& engine) const {
auto mem = engine.allocate_memory(roi_lt);
tests::set_values(mem, roi_data);
return mem;
}
void execute(const std::vector<TD>& expected_output,
roi_align::PoolingMode pooling_mode,
roi_align::AlignedMode aligned_mode) const {
auto& engine = get_test_engine();
auto input = get_memory(engine, input_lt, input_data);
auto coords = get_memory(engine, coords_lt, coords_data);
auto roi_ind = get_roi_memory(engine);
topology topology;
topology.add(input_layout("input", input->get_layout()));
topology.add(input_layout("coords", coords->get_layout()));
topology.add(input_layout("roi_ind", roi_ind->get_layout()));
topology.add(reorder("reorder_input", "input", blocked_format, device_data_type));
topology.add(reorder("reorder_coords", "coords", blocked_format, device_data_type));
topology.add(reorder("reorder_ind", "roi_ind", blocked_format, device_ind_type));
topology.add(roi_align("roi_align",
{"reorder_input", "reorder_coords", "reorder_ind"},
pooled_h,
pooled_w,
sampling_ratio,
spatial_scale,
pooling_mode,
aligned_mode));
topology.add(reorder("out", "roi_align", plain_format, device_data_type));
network network(engine, topology);
network.set_input_data("input", input);
network.set_input_data("coords", coords);
network.set_input_data("roi_ind", roi_ind);
auto outputs = network.execute();
auto output = outputs.at("out").get_memory();
cldnn::mem_lock<TD> output_ptr(output, get_test_stream());
ASSERT_EQ(output_ptr.size(), expected_output.size());
for (uint32_t i = 0; i < expected_output.size(); ++i) {
EXPECT_NEAR(output_ptr[i], expected_output[i], 0.01);
}
}
};
// it's a bit overloaded with the cartesian product of types and formats, but that's the lesser evil
// since we have specific type for expected values that are tied to specific input modes
// so that Combine approach could avoid manual combinations but it would be much more complicated
using roi_align_test_types = testing::Types<TypesWithFormat<half_t, uint8_t, format::bfyx>,
TypesWithFormat<half_t, uint8_t, format::b_fs_yx_fsv16>,
TypesWithFormat<half_t, uint8_t, format::b_fs_yx_fsv32>,
TypesWithFormat<half_t, uint8_t, format::bs_fs_yx_bsv16_fsv16>,
TypesWithFormat<half_t, uint8_t, format::bs_fs_yx_bsv32_fsv16>,
TypesWithFormat<half_t, uint8_t, format::bs_fs_yx_bsv32_fsv32>,
TypesWithFormat<half_t, int8_t, format::bfyx>,
TypesWithFormat<half_t, int8_t, format::b_fs_yx_fsv16>,
TypesWithFormat<half_t, int8_t, format::b_fs_yx_fsv32>,
TypesWithFormat<half_t, int8_t, format::bs_fs_yx_bsv16_fsv16>,
TypesWithFormat<half_t, int8_t, format::bs_fs_yx_bsv32_fsv16>,
TypesWithFormat<half_t, int8_t, format::bs_fs_yx_bsv32_fsv32>,
TypesWithFormat<half_t, int32_t, format::bfyx>,
TypesWithFormat<half_t, int32_t, format::b_fs_yx_fsv16>,
TypesWithFormat<half_t, int32_t, format::b_fs_yx_fsv32>,
TypesWithFormat<half_t, int32_t, format::bs_fs_yx_bsv16_fsv16>,
TypesWithFormat<half_t, int32_t, format::bs_fs_yx_bsv32_fsv16>,
TypesWithFormat<half_t, int32_t, format::bs_fs_yx_bsv32_fsv32>,
TypesWithFormat<float, uint8_t, format::bfyx>,
TypesWithFormat<float, uint8_t, format::b_fs_yx_fsv16>,
TypesWithFormat<float, uint8_t, format::b_fs_yx_fsv32>,
TypesWithFormat<float, uint8_t, format::bs_fs_yx_bsv16_fsv16>,
TypesWithFormat<float, uint8_t, format::bs_fs_yx_bsv32_fsv16>,
TypesWithFormat<float, uint8_t, format::bs_fs_yx_bsv32_fsv32>,
TypesWithFormat<float, int8_t, format::bfyx>,
TypesWithFormat<float, int8_t, format::b_fs_yx_fsv16>,
TypesWithFormat<float, int8_t, format::b_fs_yx_fsv32>,
TypesWithFormat<float, int8_t, format::bs_fs_yx_bsv16_fsv16>,
TypesWithFormat<float, int8_t, format::bs_fs_yx_bsv32_fsv16>,
TypesWithFormat<float, int8_t, format::bs_fs_yx_bsv32_fsv32>,
TypesWithFormat<float, int32_t, format::bfyx>,
TypesWithFormat<float, int32_t, format::b_fs_yx_fsv16>,
TypesWithFormat<float, int32_t, format::b_fs_yx_fsv32>,
TypesWithFormat<float, int32_t, format::bs_fs_yx_bsv16_fsv16>,
TypesWithFormat<float, int32_t, format::bs_fs_yx_bsv32_fsv16>,
TypesWithFormat<float, int32_t, format::bs_fs_yx_bsv32_fsv32>>;
TYPED_TEST_SUITE(roi_align_test, roi_align_test_types);
TYPED_TEST(roi_align_test, avg_asymmetric) {
using TD = typename TypeParam::DataType;
const std::vector<TD>
expected_output{TD(3.f), TD(3.75f), TD(4.75f), TD(5.f), TD(3.f), TD(5.5f), TD(2.75f), TD(3.75f)};
this->execute(expected_output, roi_align::PoolingMode::avg, roi_align::AlignedMode::asymmetric);
}
TYPED_TEST(roi_align_test, avg_half_pixel_for_nn) {
using TD = typename TypeParam::DataType;
const std::vector<TD> expected_output =
{TD(3.14f), TD(2.16f), TD(2.86f), TD(5.03f), TD(1.83f), TD(5.84f), TD(2.77f), TD(3.44f)};
this->execute(expected_output, roi_align::PoolingMode::avg, roi_align::AlignedMode::half_pixel_for_nn);
}
TYPED_TEST(roi_align_test, max_half_pixel) {
using TD = typename TypeParam::DataType;
const std::vector<TD> expected_output =
{TD(4.375f), TD(4.9375f), TD(5.6875f), TD(5.625f), TD(4.625f), TD(7.125f), TD(3.3125f), TD(4.3125f)};
this->execute(expected_output, roi_align::PoolingMode::max, roi_align::AlignedMode::half_pixel);
}

102
src/tests/functional/plugin/gpu/shared_tests_instances/single_layer_tests/roi_align.cpp
View File

@ -1,16 +1,16 @@
// Copyright (C) 2018-2022 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
//
#include "single_layer_tests/roi_align.hpp"
#include <vector>
#include "single_layer_tests/roi_align.hpp"
#include "common_test_utils/test_constants.hpp"
using namespace LayerTestsDefinitions;
const std::vector<InferenceEngine::Precision> netPRCs = {
InferenceEngine::Precision::FP32,
InferenceEngine::Precision::FP32
// There is no possibility to test ROIAlign in fp16 precision,
// because on edge cases where in fp32 version ROI value is
// a little bit smaller than the nearest integer value,
@ -19,44 +19,88 @@ const std::vector<InferenceEngine::Precision> netPRCs = {
// in fp32 and fp16 precisions.
// In real AI applications this problem is solved by precision-aware training.
// InferenceEngine::Precision::FP16,
// InferenceEngine::Precision::FP16
};
const auto ROIAlignCases_average =
::testing::Combine(
::testing::ValuesIn(
std::vector<std::vector<size_t>> {
{ 3, 8, 16, 16 },
{ 2, 1, 16, 16 },
{ 2, 1, 8, 16 }}),
::testing::Values(std::vector<size_t>{ 2, 4 }),
const auto ROIAlignCases_average = ::testing::Combine(
::testing::ValuesIn(std::vector<std::vector<size_t>>{{3, 8, 16, 16}, {2, 1, 16, 16}, {2, 1, 8, 16}}),
::testing::Values(std::vector<size_t>{2, 4}),
::testing::Values(2),
::testing::Values(2),
::testing::ValuesIn(std::vector<float> { 1, 0.625 }),
::testing::ValuesIn(std::vector<float>{1, 0.625}),
::testing::Values(2),
::testing::Values("avg"),
::testing::ValuesIn(netPRCs),
::testing::Values(CommonTestUtils::DEVICE_GPU)
);
::testing::Values(CommonTestUtils::DEVICE_GPU));
INSTANTIATE_TEST_SUITE_P(smoke_TestsROIAlign_average, ROIAlignLayerTest, ROIAlignCases_average, ROIAlignLayerTest::getTestCaseName);
INSTANTIATE_TEST_SUITE_P(smoke_TestsROIAlign_average,
ROIAlignLayerTest,
ROIAlignCases_average,
ROIAlignLayerTest::getTestCaseName);
const auto ROIAlignCases_max =
::testing::Combine(
::testing::ValuesIn(
std::vector<std::vector<size_t>> {
{ 2, 8, 20, 20 },
{ 2, 1, 20, 20 },
{ 2, 1, 10, 20 }
}),
::testing::Values(std::vector<size_t>{ 2, 4 }),
const auto ROIAlignCases_max = ::testing::Combine(
::testing::ValuesIn(std::vector<std::vector<size_t>>{{2, 8, 20, 20}, {2, 1, 20, 20}, {2, 1, 10, 20}}),
::testing::Values(std::vector<size_t>{2, 4}),
::testing::Values(2),
::testing::Values(2),
::testing::ValuesIn(std::vector<float> { 1, 0.625 }),
::testing::ValuesIn(std::vector<float>{1, 0.625}),
::testing::Values(2),
::testing::Values("max"),
::testing::ValuesIn(netPRCs),
::testing::Values(CommonTestUtils::DEVICE_GPU)
);
::testing::Values(CommonTestUtils::DEVICE_GPU));
INSTANTIATE_TEST_SUITE_P(smoke_TestsROIAlign_max, ROIAlignLayerTest, ROIAlignCases_max, ROIAlignLayerTest::getTestCaseName);
INSTANTIATE_TEST_SUITE_P(smoke_TestsROIAlign_max,
ROIAlignLayerTest,
ROIAlignCases_max,
ROIAlignLayerTest::getTestCaseName);
const auto ROIAlignCases_average_asymmetric = ::testing::Combine(
::testing::ValuesIn(std::vector<std::vector<size_t>>{{2, 1, 8, 8}, {2, 8, 20, 20}, {2, 1, 20, 20}, {2, 1, 10, 20}}),
::testing::Values(std::vector<size_t>{2, 4}),
::testing::Values(2),
::testing::Values(2),
::testing::ValuesIn(std::vector<float>{1, 0.625}),
::testing::Values(2),
::testing::Values("avg"),
::testing::Values("asymmetric"),
::testing::ValuesIn(netPRCs),
::testing::Values(CommonTestUtils::DEVICE_GPU));
INSTANTIATE_TEST_SUITE_P(smoke_TestsROIAlign_avg_asym,
ROIAlignV9LayerTest,
ROIAlignCases_average_asymmetric,
ROIAlignV9LayerTest::getTestCaseName);
const auto ROIAlignCases_average_half_pixel_for_nn = ::testing::Combine(
::testing::ValuesIn(std::vector<std::vector<size_t>>{{2, 1, 8, 8}, {2, 8, 20, 20}, {2, 1, 20, 20}, {2, 1, 10, 20}}),
::testing::Values(std::vector<size_t>{2, 4}),
::testing::Values(2),
::testing::Values(2),
::testing::ValuesIn(std::vector<float>{1, 0.625}),
::testing::Values(2),
::testing::Values("avg"),
::testing::Values("half_pixel_for_nn"),
::testing::ValuesIn(netPRCs),
::testing::Values(CommonTestUtils::DEVICE_GPU));
INSTANTIATE_TEST_SUITE_P(smoke_TestsROIAlign_avg_hpfn,
ROIAlignV9LayerTest,
ROIAlignCases_average_half_pixel_for_nn,
ROIAlignV9LayerTest::getTestCaseName);
const auto ROIAlignCases_maximum_half_pixel = ::testing::Combine(
::testing::ValuesIn(std::vector<std::vector<size_t>>{{2, 1, 8, 8}, {2, 8, 20, 20}, {2, 1, 20, 20}, {2, 1, 10, 20}}),
::testing::Values(std::vector<size_t>{2, 4}),
::testing::Values(2),
::testing::Values(2),
::testing::ValuesIn(std::vector<float>{1, 0.625}),
::testing::Values(2),
::testing::Values("max"),
::testing::Values("half_pixel"),
::testing::ValuesIn(netPRCs),
::testing::Values(CommonTestUtils::DEVICE_GPU));
INSTANTIATE_TEST_SUITE_P(smoke_TestsROIAlign_max_hp,
ROIAlignV9LayerTest,
ROIAlignCases_maximum_half_pixel,
ROIAlignV9LayerTest::getTestCaseName);

4
src/tests/functional/plugin/shared/include/single_layer_tests/roi_align.hpp
View File

@ -12,4 +12,8 @@ TEST_P(ROIAlignLayerTest, CompareWithRefs) {
Run();
}
TEST_P(ROIAlignV9LayerTest, CompareWithRefs) {
Run();
}
} // namespace LayerTestsDefinitions

28
src/tests/functional/shared_test_classes/include/shared_test_classes/single_layer/roi_align.hpp
View File

@ -37,4 +37,32 @@ private:
std::string poolingMode;
};
using roialignV9Params = std::tuple<std::vector<size_t>, // feature map shape
std::vector<size_t>, // proposal coords shape
int, // bin's row count
int, // bin's column count
float, // spatial scale
int, // pooling ratio
std::string, // pooling mode
std::string, // roi aligned mode
InferenceEngine::Precision, // net precision
LayerTestsUtils::TargetDevice>; // device name
class ROIAlignV9LayerTest : public testing::WithParamInterface<roialignV9Params>,
virtual public LayerTestsUtils::LayerTestsCommon {
public:
static std::string getTestCaseName(const testing::TestParamInfo<roialignV9Params>& obj);
protected:
void SetUp() override;
private:
int pooledH;
int pooledW;
float spatialScale;
int poolingRatio;
std::string poolingMode;
std::string roiAlignedMode;
};
} // namespace LayerTestsDefinitions

118
src/tests/functional/shared_test_classes/src/single_layer/roi_align.cpp
View File

@ -2,10 +2,13 @@
// SPDX-License-Identifier: Apache-2.0
//
#include "shared_test_classes/single_layer/roi_align.hpp"
#include <ngraph/opsets/opset3.hpp>
#include <ngraph/opsets/opset9.hpp>
#include "ngraph_functions/builders.hpp"
#include "shared_test_classes/single_layer/roi_align.hpp"
#include "openvino/core/enum_names.hpp"
using namespace InferenceEngine;
using namespace FuncTestUtils::PrecisionUtils;
@ -77,28 +80,33 @@ void ROIAlignLayerTest::SetUp() {
std::vector<size_t> inputShape;
std::vector<size_t> coordsShape;
InferenceEngine::Precision netPrecision;
std::tie(inputShape, coordsShape, pooledH, pooledW,
spatialScale, poolingRatio, poolingMode, netPrecision, targetDevice) = this->GetParam();
std::tie(inputShape,
coordsShape,
pooledH,
pooledW,
spatialScale,
poolingRatio,
poolingMode,
netPrecision,
targetDevice) = this->GetParam();
auto ngPrc = FuncTestUtils::PrecisionUtils::convertIE2nGraphPrc(netPrecision);
auto params = ngraph::builder::makeParams(ngPrc, {inputShape});
auto paramOuts = ngraph::helpers::convert2OutputVector(
ngraph::helpers::castOps2Nodes<ngraph::op::Parameter>(params));
auto paramOuts =
ngraph::helpers::convert2OutputVector(ngraph::helpers::castOps2Nodes<ngraph::op::Parameter>(params));
std::vector<float> proposalVector;
std::vector<int> roiIdxVector;
proposalVector.resize(coordsShape[0] * 4);
roiIdxVector.resize(coordsShape[0]);
fillCoordTensor(proposalVector, inputShape[2], inputShape[3],
spatialScale, poolingRatio, pooledH, pooledW);
fillCoordTensor(proposalVector, inputShape[2], inputShape[3], spatialScale, poolingRatio, pooledH, pooledW);
fillIdxTensor(roiIdxVector, inputShape[0]);
ngraph::Shape idxShape = { coordsShape[0] };
ngraph::Shape idxShape = {coordsShape[0]};
auto coords = std::make_shared<ngraph::opset1::Constant>(ngPrc, coordsShape, proposalVector.data());
auto roisIdx = std::make_shared<ngraph::opset1::Constant>(ngraph::element::i32, idxShape, roiIdxVector.data());
std::shared_ptr<ngraph::Node> roiAlign =
std::make_shared<ngraph::opset3::ROIAlign>(paramOuts[0],
std::shared_ptr<ngraph::Node> roiAlign = std::make_shared<ngraph::opset3::ROIAlign>(paramOuts[0],
coords,
roisIdx,
pooledH,
@ -109,4 +117,94 @@ void ROIAlignLayerTest::SetUp() {
ngraph::ResultVector results{std::make_shared<ngraph::opset3::Result>(roiAlign)};
function = std::make_shared<ngraph::Function>(results, params, "roi_align");
}
std::string ROIAlignV9LayerTest::getTestCaseName(const testing::TestParamInfo<roialignV9Params>& obj) {
std::vector<size_t> inputShape;
std::vector<size_t> coordsShape;
int pooledH;
int pooledW;
float spatialScale;
int poolingRatio;
std::string poolingMode;
std::string roiAlignedMode;
InferenceEngine::Precision netPrecision;
std::string targetDevice;
std::tie(inputShape,
coordsShape,
pooledH,
pooledW,
spatialScale,
poolingRatio,
poolingMode,
roiAlignedMode,
netPrecision,
targetDevice) = obj.param;
std::ostringstream result;
result << "in_shape=" << CommonTestUtils::vec2str(inputShape) << "_";
result << "coord_shape=" << CommonTestUtils::vec2str(coordsShape) << "_";
result << "pooled_h=" << pooledH << "_";
result << "pooled_w=" << pooledW << "_";
result << "spatial_scale=" << spatialScale << "_";
result << "pooling_ratio=" << poolingRatio << "_";
result << "mode=" << poolingMode << "_";
result << "mode=" << roiAlignedMode << "_";
result << "prec=" << netPrecision.name() << "_";
result << "dev=" << targetDevice;
return result.str();
}
void ROIAlignV9LayerTest::SetUp() {
std::vector<size_t> inputShape;
std::vector<size_t> coordsShape;
InferenceEngine::Precision netPrecision;
std::tie(inputShape,
coordsShape,
pooledH,
pooledW,
spatialScale,
poolingRatio,
poolingMode,
roiAlignedMode,
netPrecision,
targetDevice) = this->GetParam();
auto ngPrc = FuncTestUtils::PrecisionUtils::convertIE2nGraphPrc(netPrecision);
auto params = ngraph::builder::makeParams(ngPrc, {inputShape});
auto paramOuts =
ngraph::helpers::convert2OutputVector(ngraph::helpers::castOps2Nodes<ngraph::op::Parameter>(params));
std::vector<float> proposalVector;
std::vector<int> roiIdxVector;
proposalVector.resize(coordsShape[0] * 4);
roiIdxVector.resize(coordsShape[0]);
ROIAlignLayerTest::fillCoordTensor(proposalVector,
inputShape[2],
inputShape[3],
spatialScale,
poolingRatio,
pooledH,
pooledW);
ROIAlignLayerTest::fillIdxTensor(roiIdxVector, inputShape[0]);
ngraph::Shape idxShape = {coordsShape[0]};
auto coords = std::make_shared<ngraph::opset1::Constant>(ngPrc, coordsShape, proposalVector.data());
auto roisIdx = std::make_shared<ngraph::opset1::Constant>(ngraph::element::i32, idxShape, roiIdxVector.data());
std::shared_ptr<ngraph::Node> roiAlign = std::make_shared<ngraph::opset9::ROIAlign>(
paramOuts[0],
coords,
roisIdx,
pooledH,
pooledW,
poolingRatio,
spatialScale,
ov::EnumNames<ngraph::opset9::ROIAlign::PoolingMode>::as_enum(poolingMode),
ov::EnumNames<ngraph::opset9::ROIAlign::AlignedMode>::as_enum(roiAlignedMode));
ngraph::ResultVector results{std::make_shared<ngraph::opset9::Result>(roiAlign)};
function = std::make_shared<ngraph::Function>(results, params, "roi_align");
}
} // namespace LayerTestsDefinitions