Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[FrontEnd]enable pdpd ops conversion part4: deformable_convolution, multiclass_nms and matrix_nms. #6833

Merged
merged 1 commit into from
Aug 3, 2021
Merged

[FrontEnd]enable pdpd ops conversion part4: deformable_convolution, multiclass_nms and matrix_nms. #6833

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
74 changes: 74 additions & 0 deletions ngraph/frontend/paddlepaddle/src/op/deformable_conv.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,74 @@
// Copyright (C) 2018-2021 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
//
#include <ngraph/opsets/opset8.hpp>
#include <node_context.hpp>
#include "conv2d_utils.hpp"

namespace ngraph
{
namespace frontend
{
namespace pdpd
{
namespace op
{
NamedOutputs deformable_conv(const NodeContext& node)
{
auto input = node.get_ng_input("Input");
auto filter = node.get_ng_input("Filter");
auto offset = node.get_ng_input("Offset");

auto strides = node.get_attribute<std::vector<int>>("strides");
auto dilations = node.get_attribute<std::vector<int>>("dilations");

auto groups = node.get_attribute<int>("groups");
auto deformable_groups = node.get_attribute<int>("deformable_groups");

const auto paddings = get_pads(node);
const auto pads_begin = paddings.first;
const auto pads_end = paddings.second;

const ngraph::op::PadType auto_pad{ngraph::op::PadType::EXPLICIT};

std::shared_ptr<Node> output_node;
if (node.has_ng_input("Mask"))
{
auto mask = node.get_ng_input("Mask");
output_node = std::make_shared<ngraph::opset8::DeformableConvolution>(
input,
offset,
filter,
mask,
ngraph::Strides(strides.begin(), strides.end()),
pads_begin,
pads_end,
ngraph::Strides(dilations.begin(), dilations.end()),
auto_pad,
groups,
deformable_groups,
true);
}
else
{
output_node = std::make_shared<ngraph::opset8::DeformableConvolution>(
input,
offset,
filter,
ngraph::Strides(strides.begin(), strides.end()),
pads_begin,
pads_end,
ngraph::Strides(dilations.begin(), dilations.end()),
auto_pad,
groups,
deformable_groups,
true);
}

return node.default_single_output_mapping({output_node}, {"Output"});
}

} // namespace op
} // namespace pdpd
} // namespace frontend
} // namespace ngraph
99 changes: 99 additions & 0 deletions ngraph/frontend/paddlepaddle/src/op/matrix_nms.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,99 @@
// Copyright (C) 2018-2021 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
//
#include <ngraph/opsets/opset8.hpp>
#include <node_context.hpp>
#include <paddlepaddle_frontend/utility.hpp>

namespace ngraph
{
namespace frontend
{
namespace pdpd
{
namespace op
{
NamedOutputs matrix_nms(const NodeContext& node)
{
using namespace ngraph;
using namespace opset8;
using namespace element;

auto bboxes = node.get_ng_input("BBoxes");
auto scores = node.get_ng_input("Scores");

auto score_threshold = node.get_attribute<float>("score_threshold");
auto post_threshold = node.get_attribute<float>("post_threshold");
auto nms_top_k = node.get_attribute<int>("nms_top_k");
auto keep_top_k = node.get_attribute<int>("keep_top_k");
auto background_class = node.get_attribute<int>("background_label");

auto gaussian_sigma = node.get_attribute<float>("gaussian_sigma");
auto use_gaussian = node.get_attribute<bool>("use_gaussian");
auto decay_function = MatrixNms::DecayFunction::LINEAR;
if (use_gaussian)
{
decay_function = MatrixNms::DecayFunction::GAUSSIAN;
}

auto out_names = node.get_output_names();
PDPD_ASSERT(out_names.size() == 3 || out_names.size() == 2,
"Unexpected number of outputs of MatrixNMS: " + out_names.size());

element::Type type_num = i32;
bool return_rois_num = true;
auto it = std::find(out_names.begin(), out_names.end(), "RoisNum");
if (it != out_names.end())
{
type_num = node.get_out_port_type("RoisNum");
}
else
{
return_rois_num = false;
}

auto type_index = node.get_out_port_type("Index");
PDPD_ASSERT((type_index == i32 || type_index == i64) &&
(type_num == i32 || type_num == i64),
"Unexpected data type of outputs of MatrixNMS");

auto normalized = node.get_attribute<bool>("normalized");

NamedOutputs named_outputs;
std::vector<Output<Node>> nms_outputs;
MatrixNms::Attributes attrs;
attrs.nms_top_k = nms_top_k;
attrs.post_threshold = post_threshold;
attrs.score_threshold = score_threshold;
attrs.sort_result_type = MatrixNms::SortResultType::SCORE;
attrs.keep_top_k = keep_top_k;
attrs.background_class = background_class;
attrs.normalized = normalized;
attrs.output_type = type_index;
attrs.sort_result_across_batch = false;
attrs.decay_function = decay_function;
attrs.gaussian_sigma = gaussian_sigma;

nms_outputs = std::make_shared<MatrixNms>(bboxes, scores, attrs)->outputs();

named_outputs["Out"] = {nms_outputs[0]};
named_outputs["Index"] = {nms_outputs[1]};
if (return_rois_num)
{
named_outputs["RoisNum"] = {nms_outputs[2]};

if (type_num != type_index)
{
// adapter
auto node_convert = std::make_shared<Convert>(nms_outputs[2], type_num);
named_outputs["RoisNum"] = {node_convert};
}
}

return named_outputs;
}

} // namespace op
} // namespace pdpd
} // namespace frontend
} // namespace ngraph
78 changes: 78 additions & 0 deletions ngraph/frontend/paddlepaddle/src/op/multiclass_nms.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,78 @@
// Copyright (C) 2018-2021 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
//
#include <ngraph/opsets/opset8.hpp>
#include <node_context.hpp>
#include <paddlepaddle_frontend/utility.hpp>

namespace ngraph
{
namespace frontend
{
namespace pdpd
{
namespace op
{
NamedOutputs multiclass_nms(const NodeContext& node)
{
using namespace ngraph;
using namespace opset8;
using namespace element;

auto bboxes = node.get_ng_input("BBoxes");
auto scores = node.get_ng_input("Scores");

auto score_threshold = node.get_attribute<float>("score_threshold");
auto iou_threshold = node.get_attribute<float>("nms_threshold");
auto nms_top_k = node.get_attribute<int>("nms_top_k");
auto keep_top_k = node.get_attribute<int>("keep_top_k");
auto background_class = node.get_attribute<int>("background_label");
auto nms_eta = node.get_attribute<float>("nms_eta");

auto out_names = node.get_output_names();
PDPD_ASSERT(out_names.size() == 3,
"Unexpected number of outputs of MulticlassNMS");

auto type_index = node.get_out_port_type("Index");
auto type_num = node.get_out_port_type("NmsRoisNum");
PDPD_ASSERT((type_index == i32 || type_index == i64) &&
(type_num == i32 || type_num == i64),
"Unexpected data type of outputs of MulticlassNMS: " +
out_names.size());

auto normalized = node.get_attribute<bool>("normalized");

NamedOutputs named_outputs;
std::vector<Output<Node>> nms_outputs;
MulticlassNms::Attributes attrs;
attrs.nms_top_k = nms_top_k;
attrs.iou_threshold = iou_threshold;
attrs.score_threshold = score_threshold;
attrs.sort_result_type = MulticlassNms::SortResultType::CLASSID;
attrs.keep_top_k = keep_top_k;
attrs.background_class = background_class;
attrs.nms_eta = nms_eta;
attrs.normalized = normalized;
attrs.output_type = type_index;
attrs.sort_result_across_batch = false;

nms_outputs = std::make_shared<MulticlassNms>(bboxes, scores, attrs)->outputs();

named_outputs["Out"] = {nms_outputs[0]};
named_outputs["Index"] = {nms_outputs[1]};
named_outputs["NmsRoisNum"] = {nms_outputs[2]};

if (type_num != type_index)
{
// adapter
auto node_convert = std::make_shared<Convert>(nms_outputs[2], type_num);
named_outputs["NmsRoisNum"] = {node_convert};
}

return named_outputs;
}

} // namespace op
} // namespace pdpd
} // namespace frontend
} // namespace ngraph
8 changes: 7 additions & 1 deletion ngraph/frontend/paddlepaddle/src/op_table.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,6 @@
// Copyright (C) 2018-2021 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
//

#include "op_table.hpp"

namespace ngraph
Expand All @@ -23,6 +22,7 @@ namespace ngraph
OP_CONVERTER(concat);
OP_CONVERTER(conv2d);
OP_CONVERTER(conv2d_transpose);
OP_CONVERTER(deformable_conv);
OP_CONVERTER(dropout);
OP_CONVERTER(elementwise_add);
OP_CONVERTER(elementwise_div);
Expand All @@ -43,6 +43,8 @@ namespace ngraph
OP_CONVERTER(log);
OP_CONVERTER(logical_not);
OP_CONVERTER(matmul);
OP_CONVERTER(matrix_nms);
OP_CONVERTER(multiclass_nms);
OP_CONVERTER(nearest_interp_v2);
OP_CONVERTER(rnn);
OP_CONVERTER(relu);
Expand Down Expand Up @@ -74,6 +76,8 @@ namespace ngraph
{"concat", op::concat},
{"conv2d", op::conv2d},
{"conv2d_transpose", op::conv2d_transpose},
{"deformable_conv", op::deformable_conv},
{"deformable_conv_v1", op::deformable_conv},
{"depthwise_conv2d", op::conv2d},
{"depthwise_conv2d_transpose", op::conv2d_transpose},
{"dropout", op::dropout},
Expand All @@ -96,6 +100,8 @@ namespace ngraph
{"log", op::log},
{"logical_not", op::logical_not},
{"matmul", op::matmul},
{"matrix_nms", op::matrix_nms},
{"multiclass_nms3", op::multiclass_nms},
{"nearest_interp_v2", op::nearest_interp_v2},
{"nearest_interp", op::nearest_interp_v2},
{"rnn", op::rnn},
Expand Down