apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-22 00:02:10 +08:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity

Add custom operator for onnxruntime and fix paddle backend (#35)

Browse Source 
Add custom operator for onnxruntime ans fix paddle backend
This commit is contained in:
Jason
2022-07-23 22:21:36 +08:00
committed by GitHub
parent 51ecb407d4
commit 4b681581b1
12 changed files with 666 additions and 60 deletions
Download Patch File Download Diff File Expand all files Collapse all files

209
ThirdPartyNotices.txt
View File

@@ -732,3 +732,212 @@ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
---------
7. https://github.com/oneapi-src/oneDNN/
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
============================================================================
Copyright 2016-2021 Intel Corporation
Copyright 2018 YANDEX LLC
Copyright 2019-2021 FUJITSU LIMITED
Copyright 2020 Arm Limited and affiliates
Copyright 2020 Codeplay Software Limited
Copyright 2021 Alanna Tempest
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
This distribution includes third party software ("third party programs").
This third party software, even if included with the distribution of
the Intel software, may be governed by separate license terms, including
without limitation, third party license terms, other Intel software license
terms, and open source software license terms. These separate license terms
govern your use of the third party programs as set forth in the
"THIRD-PARTY-PROGRAMS" file.

7
external/paddle_inference.cmake vendored
View File

@@ -83,12 +83,7 @@ ExternalProject_Add(
BUILD_COMMAND ""
UPDATE_COMMAND ""
INSTALL_COMMAND
${CMAKE_COMMAND} -E remove_directory ${PADDLEINFERENCE_INSTALL_DIR} &&
${CMAKE_COMMAND} -E make_directory ${PADDLEINFERENCE_INSTALL_DIR} &&
${CMAKE_COMMAND} -E rename ${PADDLEINFERENCE_SOURCE_DIR}/paddle/
${PADDLEINFERENCE_INSTALL_DIR}/paddle && ${CMAKE_COMMAND} -E rename
${PADDLEINFERENCE_SOURCE_DIR}/third_party ${PADDLEINFERENCE_INSTALL_DIR}/third_party &&
${CMAKE_COMMAND} -E rename ${PADDLEINFERENCE_SOURCE_DIR}/version.txt ${PADDLEINFERENCE_INSTALL_DIR}/version.txt
${CMAKE_COMMAND} -E copy_directory ${PADDLEINFERENCE_SOURCE_DIR} ${PADDLEINFERENCE_INSTALL_DIR}
BUILD_BYPRODUCTS ${PADDLEINFERENCE_COMPILE_LIB})
add_library(external_paddle_inference STATIC IMPORTED GLOBAL)

13
external/utils.cmake vendored
View File

@@ -13,3 +13,16 @@ function(redefine_file_macro targetname)
)
endforeach()
endfunction()
function(download_and_decompress url filename decompress_dir)
if(NOT EXISTS ${filename})
message("Downloading file from ${url} ...")
file(DOWNLOAD ${url} "${filename}.tmp" SHOW_PROGRESS)
file(RENAME "${filename}.tmp" ${filename})
endif()
if(NOT EXISTS ${decompress_dir})
file(MAKE_DIRECTORY ${decompress_dir})
endif()
message("Decompress file ${filename} ...")
execute_process(COMMAND ${CMAKE_COMMAND} -E tar -xf ${filename} WORKING_DIRECTORY ${decompress_dir})
endfunction()

260
fastdeploy/backends/ort/ops/multiclass_nms.cc Normal file
View File

@@ -0,0 +1,260 @@
// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "fastdeploy/backends/ort/ops/multiclass_nms.h"
#include <algorithm>
#include "fastdeploy/core/fd_tensor.h"
#include "fastdeploy/utils/utils.h"
namespace fastdeploy {
struct OrtTensorDimensions : std::vector<int64_t> {
OrtTensorDimensions(Ort::CustomOpApi ort, const OrtValue* value) {
OrtTensorTypeAndShapeInfo* info = ort.GetTensorTypeAndShape(value);
std::vector<int64_t>::operator=(ort.GetTensorShape(info));
ort.ReleaseTensorTypeAndShapeInfo(info);
}
};
template <class T>
bool SortScorePairDescend(const std::pair<float, T>& pair1,
const std::pair<float, T>& pair2) {
return pair1.first > pair2.first;
}
void GetMaxScoreIndex(const float* scores, const int& score_size,
const float& threshold, const int& top_k,
std::vector<std::pair<float, int>>* sorted_indices) {
for (size_t i = 0; i < score_size; ++i) {
if (scores[i] > threshold) {
sorted_indices->push_back(std::make_pair(scores[i], i));
}
}
// Sort the score pair according to the scores in descending order
std::stable_sort(sorted_indices->begin(), sorted_indices->end(),
SortScorePairDescend<int>);
// Keep top_k scores if needed.
if (top_k > -1 && top_k < static_cast<int>(sorted_indices->size())) {
sorted_indices->resize(top_k);
}
}
float BBoxArea(const float* box, const bool& normalized) {
if (box[2] < box[0] || box[3] < box[1]) {
// If coordinate values are is invalid
// (e.g. xmax < xmin or ymax < ymin), return 0.
return 0.f;
} else {
const float w = box[2] - box[0];
const float h = box[3] - box[1];
if (normalized) {
return w * h;
} else {
// If coordinate values are not within range [0, 1].
return (w + 1) * (h + 1);
}
}
}
float JaccardOverlap(const float* box1, const float* box2,
const bool& normalized) {
if (box2[0] > box1[2] || box2[2] < box1[0] || box2[1] > box1[3] ||
box2[3] < box1[1]) {
return 0.f;
} else {
const float inter_xmin = std::max(box1[0], box2[0]);
const float inter_ymin = std::max(box1[1], box2[1]);
const float inter_xmax = std::min(box1[2], box2[2]);
const float inter_ymax = std::min(box1[3], box2[3]);
float norm = normalized ? 0.0f : 1.0f;
float inter_w = inter_xmax - inter_xmin + norm;
float inter_h = inter_ymax - inter_ymin + norm;
const float inter_area = inter_w * inter_h;
const float bbox1_area = BBoxArea(box1, normalized);
const float bbox2_area = BBoxArea(box2, normalized);
return inter_area / (bbox1_area + bbox2_area - inter_area);
}
}
void MultiClassNmsKernel::FastNMS(const float* boxes, const float* scores,
const int& num_boxes,
std::vector<int>* keep_indices) {
std::vector<std::pair<float, int>> sorted_indices;
GetMaxScoreIndex(scores, num_boxes, score_threshold, nms_top_k,
&sorted_indices);
float adaptive_threshold = nms_threshold;
while (sorted_indices.size() != 0) {
const int idx = sorted_indices.front().second;
bool keep = true;
for (size_t k = 0; k < keep_indices->size(); ++k) {
if (!keep) {
break;
}
const int kept_idx = (*keep_indices)[k];
float overlap =
JaccardOverlap(boxes + idx * 4, boxes + kept_idx * 4, normalized);
keep = overlap <= adaptive_threshold;
}
if (keep) {
keep_indices->push_back(idx);
}
sorted_indices.erase(sorted_indices.begin());
if (keep && nms_eta<1.0 & adaptive_threshold> 0.5) {
adaptive_threshold *= nms_eta;
}
}
}
int MultiClassNmsKernel::NMSForEachSample(
const float* boxes, const float* scores, int num_boxes, int num_classes,
std::map<int, std::vector<int>>* keep_indices) {
for (int i = 0; i < num_classes; ++i) {
if (i == background_label) {
continue;
}
const float* score_for_class_i = scores + i * num_boxes;
FastNMS(boxes, score_for_class_i, num_boxes, &((*keep_indices)[i]));
}
int num_det = 0;
for (auto iter = keep_indices->begin(); iter != keep_indices->end(); ++iter) {
num_det += iter->second.size();
}
if (keep_top_k > -1 && num_det > keep_top_k) {
std::vector<std::pair<float, std::pair<int, int>>> score_index_pairs;
for (const auto& it : *keep_indices) {
int label = it.first;
const float* current_score = scores + label * num_boxes;
auto& label_indices = it.second;
for (size_t j = 0; j < label_indices.size(); ++j) {
int idx = label_indices[j];
score_index_pairs.push_back(
std::make_pair(current_score[idx], std::make_pair(label, idx)));
}
}
std::stable_sort(score_index_pairs.begin(), score_index_pairs.end(),
SortScorePairDescend<std::pair<int, int>>);
score_index_pairs.resize(keep_top_k);
std::map<int, std::vector<int>> new_indices;
for (size_t j = 0; j < score_index_pairs.size(); ++j) {
int label = score_index_pairs[j].second.first;
int idx = score_index_pairs[j].second.second;
new_indices[label].push_back(idx);
}
new_indices.swap(*keep_indices);
num_det = keep_top_k;
}
return num_det;
}
void MultiClassNmsKernel::Compute(OrtKernelContext* context) {
const OrtValue* boxes = ort_.KernelContext_GetInput(context, 0);
const OrtValue* scores = ort_.KernelContext_GetInput(context, 1);
const float* boxes_data =
reinterpret_cast<const float*>(ort_.GetTensorData<float>(boxes));
const float* scores_data =
reinterpret_cast<const float*>(ort_.GetTensorData<float>(scores));
OrtTensorDimensions boxes_dim(ort_, boxes);
OrtTensorDimensions scores_dim(ort_, scores);
int score_size = scores_dim.size();
int64_t batch_size = scores_dim[0];
int64_t box_dim = boxes_dim[2];
int64_t out_dim = box_dim + 2;
int num_nmsed_out = 0;
FDASSERT(score_size == 3, "Require rank of input scores be 3, but now it's " +
std::to_string(score_size) + ".");
FDASSERT(boxes_dim[2] == 4,
"Require the 3-dimension of input boxes be 4, but now it's " +
std::to_string(boxes_dim[2]) + ".");
std::vector<int64_t> out_num_rois_dims = {batch_size};
OrtValue* out_num_rois = ort_.KernelContext_GetOutput(
context, 2, out_num_rois_dims.data(), out_num_rois_dims.size());
int32_t* out_num_rois_data = ort_.GetTensorMutableData<int32_t>(out_num_rois);
std::vector<std::map<int, std::vector<int>>> all_indices;
for (size_t i = 0; i < batch_size; ++i) {
std::map<int, std::vector<int>> indices; // indices kept for each class
const float* current_boxes_ptr =
boxes_data + i * boxes_dim[1] * boxes_dim[2];
const float* current_scores_ptr =
scores_data + i * scores_dim[1] * scores_dim[2];
int num = NMSForEachSample(current_boxes_ptr, current_scores_ptr,
boxes_dim[1], scores_dim[1], &indices);
num_nmsed_out += num;
out_num_rois_data[i] = num;
all_indices.emplace_back(indices);
}
std::vector<int64_t> out_box_dims = {num_nmsed_out, 6};
std::vector<int64_t> out_index_dims = {num_nmsed_out, 1};
OrtValue* out_box = ort_.KernelContext_GetOutput(
context, 0, out_box_dims.data(), out_box_dims.size());
OrtValue* out_index = ort_.KernelContext_GetOutput(
context, 1, out_index_dims.data(), out_index_dims.size());
if (num_nmsed_out == 0) {
int32_t* out_num_rois_data =
ort_.GetTensorMutableData<int32_t>(out_num_rois);
for (size_t i = 0; i < batch_size; ++i) {
out_num_rois_data[i] = 0;
}
return;
}
float* out_box_data = ort_.GetTensorMutableData<float>(out_box);
int32_t* out_index_data = ort_.GetTensorMutableData<int32_t>(out_index);
int count = 0;
for (size_t i = 0; i < batch_size; ++i) {
const float* current_boxes_ptr =
boxes_data + i * boxes_dim[1] * boxes_dim[2];
const float* current_scores_ptr =
scores_data + i * scores_dim[1] * scores_dim[2];
for (const auto& it : all_indices[i]) {
int label = it.first;
const auto& indices = it.second;
const float* current_scores_class_ptr =
current_scores_ptr + label * scores_dim[2];
for (size_t j = 0; j < indices.size(); ++j) {
int start = count * 6;
out_box_data[start] = label;
out_box_data[start + 1] = current_scores_class_ptr[indices[j]];
out_box_data[start + 2] = current_boxes_ptr[indices[j] * 4];
out_box_data[start + 3] = current_boxes_ptr[indices[j] * 4 + 1];
out_box_data[start + 4] = current_boxes_ptr[indices[j] * 4 + 2];
out_box_data[start + 5] = current_boxes_ptr[indices[j] * 4 + 3];
out_index_data[count] = i * boxes_dim[1] + indices[j];
count += 1;
}
}
}
}
void MultiClassNmsKernel::GetAttribute(const OrtKernelInfo* info) {
background_label =
ort_.KernelInfoGetAttribute<int64_t>(info, "background_label");
keep_top_k = ort_.KernelInfoGetAttribute<int64_t>(info, "keep_top_k");
nms_eta = ort_.KernelInfoGetAttribute<float>(info, "nms_eta");
nms_threshold = ort_.KernelInfoGetAttribute<float>(info, "nms_threshold");
nms_top_k = ort_.KernelInfoGetAttribute<int64_t>(info, "nms_top_k");
normalized = ort_.KernelInfoGetAttribute<int64_t>(info, "normalized");
score_threshold = ort_.KernelInfoGetAttribute<float>(info, "score_threshold");
std::cout << background_label << " " << keep_top_k << " " << nms_eta << " "
<< nms_threshold << " " << nms_top_k << " " << normalized << " "
<< score_threshold << " " << std::endl;
}
} // namespace fastdeploy

76
fastdeploy/backends/ort/ops/multiclass_nms.h Normal file
View File

@@ -0,0 +1,76 @@
// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#pragma once
#include <map>
#include "onnxruntime_cxx_api.h" // NOLINT
namespace fastdeploy {
struct MultiClassNmsKernel {
protected:
int64_t background_label = -1;
int64_t keep_top_k = -1;
float nms_eta;
float nms_threshold = 0.7;
int64_t nms_top_k;
bool normalized;
float score_threshold;
Ort::CustomOpApi ort_;
public:
MultiClassNmsKernel(Ort::CustomOpApi ort, const OrtKernelInfo* info)
: ort_(ort) {
GetAttribute(info);
}
void GetAttribute(const OrtKernelInfo* info);
void Compute(OrtKernelContext* context);
void FastNMS(const float* boxes, const float* scores, const int& num_boxes,
std::vector<int>* keep_indices);
int NMSForEachSample(const float* boxes, const float* scores, int num_boxes,
int num_classes,
std::map<int, std::vector<int>>* keep_indices);
};
struct MultiClassNmsOp
: Ort::CustomOpBase<MultiClassNmsOp, MultiClassNmsKernel> {
void* CreateKernel(Ort::CustomOpApi api, const OrtKernelInfo* info) const {
return new MultiClassNmsKernel(api, info);
}
const char* GetName() const { return "MultiClassNMS"; }
size_t GetInputTypeCount() const { return 2; }
ONNXTensorElementDataType GetInputType(size_t index) const {
return ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT;
}
size_t GetOutputTypeCount() const { return 3; }
ONNXTensorElementDataType GetOutputType(size_t index) const {
if (index == 0) {
return ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT;
}
return ONNX_TENSOR_ELEMENT_DATA_TYPE_INT32;
}
const char* GetExecutionProviderType() const {
return "CPUExecutionProvider";
}
};
} // namespace fastdeploy

19
fastdeploy/backends/ort/ort_backend.cc
View File

@@ -13,15 +13,19 @@
// limitations under the License.
#include "fastdeploy/backends/ort/ort_backend.h"
#include <memory>
#include "fastdeploy/backends/ort/ops/multiclass_nms.h"
#include "fastdeploy/backends/ort/utils.h"
#include "fastdeploy/utils/utils.h"
#include <memory>
#ifdef ENABLE_PADDLE_FRONTEND
#include "paddle2onnx/converter.h"
#endif
namespace fastdeploy {
std::vector<OrtCustomOp*> OrtBackend::custom_operators_ =
std::vector<OrtCustomOp*>();
ONNXTensorElementDataType GetOrtDtype(FDDataType fd_dtype) {
if (fd_dtype == FDDataType::FP32) {
return ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT;
@@ -131,7 +135,9 @@ bool OrtBackend::InitFromOnnx(const std::string& model_file,
<< std::endl;
return false;
}
BuildOption(option);
InitCustomOperators();
if (from_memory_buffer) {
session_ = {env_, model_file.data(), model_file.size(), session_options_};
} else {
@@ -275,4 +281,15 @@ TensorInfo OrtBackend::GetOutputInfo(int index) {
return info;
}
void OrtBackend::InitCustomOperators() {
if (custom_operators_.size() == 0) {
MultiClassNmsOp* custom_op = new MultiClassNmsOp{};
custom_operators_.push_back(custom_op);
}
for (size_t i = 0; i < custom_operators_.size(); ++i) {
custom_op_domain_.Add(custom_operators_[i]);
}
session_options_.Add(custom_op_domain_);
}
} // namespace fastdeploy

5
fastdeploy/backends/ort/ort_backend.h
View File

@@ -68,6 +68,8 @@ class OrtBackend : public BaseBackend {
TensorInfo GetInputInfo(int index);
TensorInfo GetOutputInfo(int index);
static std::vector<OrtCustomOp*> custom_operators_;
void InitCustomOperators();
private:
Ort::Env env_;
@@ -76,9 +78,8 @@ class OrtBackend : public BaseBackend {
std::shared_ptr<Ort::IoBinding> binding_;
std::vector<OrtValueInfo> inputs_desc_;
std::vector<OrtValueInfo> outputs_desc_;
Ort::CustomOpDomain custom_op_domain_ = Ort::CustomOpDomain("Paddle");
OrtBackendOption option_;
void CopyToCpu(const Ort::Value& value, FDTensor* tensor);
};
} // namespace fastdeploy

1
fastdeploy/backends/paddle/util.cc
View File

@@ -17,6 +17,7 @@
namespace fastdeploy {
void ShareTensorFromCpu(paddle_infer::Tensor* tensor, FDTensor& fd_tensor) {
std::vector<int> shape(fd_tensor.shape.begin(), fd_tensor.shape.end());
tensor->Reshape(shape);
if (fd_tensor.dtype == FDDataType::FP32) {
tensor->ShareExternalData(static_cast<const float*>(fd_tensor.Data()),
shape, paddle_infer::PlaceType::kCPU);

2
fastdeploy/fastdeploy_model.cc
View File

@@ -18,7 +18,7 @@ namespace fastdeploy {
bool FastDeployModel::InitRuntime() {
FDASSERT(
ModelFormatCheck(runtime_option.model_file, runtime_option.model_format),
CheckModelFormat(runtime_option.model_file, runtime_option.model_format),
"ModelFormatCheck Failed.");
if (runtime_initialized_) {
FDERROR << "The model is already initialized, cannot be initliazed again."

29
fastdeploy/fastdeploy_runtime.cc
View File

@@ -72,7 +72,7 @@ std::string Str(const Frontend& f) {
return "UNKNOWN-Frontend";
}
bool ModelFormatCheck(const std::string& model_file,
bool CheckModelFormat(const std::string& model_file,
const Frontend& model_format) {
if (model_format == Frontend::PADDLE) {
if (model_file.size() < 8 ||
@@ -99,8 +99,28 @@ bool ModelFormatCheck(const std::string& model_file,
return true;
}
Frontend GuessModelFormat(const std::string& model_file) {
if (model_file.size() > 8 &&
model_file.substr(model_file.size() - 8, 8) == ".pdmodel") {
FDLogger() << "Model Format: PaddlePaddle." << std::endl;
return Frontend::PADDLE;
} else if (model_file.size() > 5 &&
model_file.substr(model_file.size() - 5, 5) == ".onnx") {
FDLogger() << "Model Format: ONNX." << std::endl;
return Frontend::ONNX;
}
FDERROR << "Cannot guess which model format you are using, please set "
"RuntimeOption::model_format manually."
<< std::endl;
return Frontend::PADDLE;
}
bool Runtime::Init(const RuntimeOption& _option) {
option = _option;
if (option.model_format == Frontend::AUTOREC) {
option.model_format = GuessModelFormat(_option.model_file);
}
if (option.backend == Backend::UNKNOWN) {
if (IsBackendAvailable(Backend::ORT)) {
option.backend = Backend::ORT;
@@ -124,6 +144,9 @@ bool Runtime::Init(const RuntimeOption& _option) {
} else if (option.backend == Backend::PDINFER) {
FDASSERT(option.device == Device::CPU || option.device == Device::GPU,
"Backend::TRT only supports Device::CPU/Device::GPU.");
FDASSERT(
option.model_format == Frontend::PADDLE,
"Backend::PDINFER only supports model format of Frontend::PADDLE.");
CreatePaddleBackend();
} else {
FDERROR << "Runtime only support "
@@ -163,8 +186,8 @@ void Runtime::CreatePaddleBackend() {
"Load model from Paddle failed while initliazing PaddleBackend.");
#else
FDASSERT(false,
"OrtBackend is not available, please compiled with "
"ENABLE_ORT_BACKEND=ON.");
"PaddleBackend is not available, please compiled with "
"ENABLE_PADDLE_BACKEND=ON.");
#endif
}

9
fastdeploy/fastdeploy_runtime.h
View File

@@ -21,7 +21,9 @@
namespace fastdeploy {
enum FASTDEPLOY_DECL Backend { UNKNOWN, ORT, TRT, PDINFER };
enum FASTDEPLOY_DECL Frontend { PADDLE, ONNX };
// AUTOREC will according to the name of model file
// to decide which Frontend is
enum FASTDEPLOY_DECL Frontend { AUTOREC, PADDLE, ONNX };
FASTDEPLOY_DECL std::string Str(const Backend& b);
FASTDEPLOY_DECL std::string Str(const Frontend& f);
@@ -29,8 +31,9 @@ FASTDEPLOY_DECL std::vector<Backend> GetAvailableBackends();
FASTDEPLOY_DECL bool IsBackendAvailable(const Backend& backend);
bool ModelFormatCheck(const std::string& model_file,
bool CheckModelFormat(const std::string& model_file,
const Frontend& model_format);
Frontend GuessModelFormat(const std::string& model_file);
struct FASTDEPLOY_DECL RuntimeOption {
Backend backend = Backend::UNKNOWN;
@@ -71,7 +74,7 @@ struct FASTDEPLOY_DECL RuntimeOption {
std::string model_file = ""; // Path of model file
std::string params_file = ""; // Path of parameters file, can be empty
Frontend model_format = Frontend::PADDLE; // format of input model
Frontend model_format = Frontend::AUTOREC; // format of input model
};
struct FASTDEPLOY_DECL Runtime {

90
setup.py
View File

@@ -126,6 +126,15 @@ class ONNXCommand(setuptools.Command):
pass
def GetAllFiles(dirname):
files = list()
for root, dirs, filenames in os.walk(dirname):
for f in filenames:
fullname = os.path.join(root, f)
files.append(fullname)
return files
class create_version(ONNXCommand):
def run(self):
with open(os.path.join(SRC_DIR, 'version.py'), 'w') as f:
@@ -326,50 +335,49 @@ if sys.argv[1] == "install" or sys.argv[1] == "bdist_wheel":
shutil.copy("LICENSE", "fastdeploy")
depend_libs = list()
if platform.system().lower() == "linux":
for f in os.listdir(".setuptools-cmake-build"):
full_name = os.path.join(".setuptools-cmake-build", f)
if not os.path.isfile(full_name):
continue
if not full_name.count("fastdeploy_main.cpython-"):
continue
if not full_name.endswith(".so"):
continue
# modify the search path of libraries
command = "patchelf --set-rpath '$ORIGIN/libs/' {}".format(
full_name)
# The sw_64 not suppot patchelf, so we just disable that.
if platform.machine() != 'sw_64' and platform.machine(
) != 'mips64':
assert os.system(
command
) == 0, "patch fastdeploy_main.cpython-36m-x86_64-linux-gnu.so failed, the command: {}".format(
command)
# copy fastdeploy library
pybind_so_file = None
for f in os.listdir(".setuptools-cmake-build"):
if not os.path.isfile(os.path.join(".setuptools-cmake-build", f)):
continue
if f.count("libfastdeploy") > 0:
if f.count("fastdeploy") > 0:
shutil.copy(
os.path.join(".setuptools-cmake-build", f), "fastdeploy/libs")
for dirname in os.listdir(".setuptools-cmake-build/third_libs/install"):
for lib in os.listdir(
os.path.join(".setuptools-cmake-build/third_libs/install",
dirname, "lib")):
if lib.count(".so") == 0 and lib.count(
".dylib") == 0 and lib.count(".a") == 0:
continue
if not os.path.isfile(
os.path.join(".setuptools-cmake-build/third_libs/install",
dirname, "lib", lib)):
continue
shutil.copy(
os.path.join(".setuptools-cmake-build/third_libs/install",
dirname, "lib", lib), "fastdeploy/libs")
if f.count("fastdeploy_main.cpython-"):
pybind_so_file = f
all_libs = os.listdir("fastdeploy/libs")
for lib in all_libs:
package_data[PACKAGE_NAME].append(os.path.join("libs", lib))
if not os.path.exists(".setuptools-cmake-build/third_libs/install"):
raise Exception(
"Cannot find directory third_libs/install in .setuptools-cmake-build."
)
if os.path.exists("fastdeploy/libs/third_libs"):
shutil.rmtree("fastdeploy/libs/third_libs")
shutil.copytree(
".setuptools-cmake-build/third_libs/install",
"fastdeploy/libs/third_libs",
symlinks=True)
all_files = GetAllFiles("fastdeploy/libs")
for f in all_files:
package_data[PACKAGE_NAME].append(os.path.relpath(f, "fastdeploy"))
if platform.system().lower() == "linux":
rpaths = ["${ORIGIN}"]
for root, dirs, files in os.walk("fastdeploy/libs/third_libs"):
for d in dirs:
if d == "lib":
path = os.path.relpath(
os.path.join(root, d), "fastdeploy/libs")
rpaths.append("${ORIGIN}/" + format(path))
rpaths = ":".join(rpaths)
command = "patchelf --set-rpath '{}' ".format(rpaths) + os.path.join(
"fastdeploy/libs", pybind_so_file)
# The sw_64 not suppot patchelf, so we just disable that.
if platform.machine() != 'sw_64' and platform.machine() != 'mips64':
assert os.system(
command) == 0, "patchelf {} failed, the command: {}".format(
command, pybind_so_file)
setuptools.setup(
name=PACKAGE_NAME,
@@ -382,9 +390,9 @@ setuptools.setup(
include_package_data=True,
setup_requires=setup_requires,
extras_require=extras_require,
author='paddle-infer',
author_email='paddle-infer@baidu.com',
url='https://github.com/PaddlePaddle/Paddle2ONNX.git',
author='fastdeploy',
author_email='fastdeploy@baidu.com',
url='https://github.com/PaddlePaddle/FastDeploy.git',
install_requires=REQUIRED_PACKAGES,
classifiers=[
"Programming Language :: Python :: 3",