init commit

README.md

-# convert_onnx
 
-examples of converting caffe, pytorch, tf, tflite, paddle model to onnx model.
+# onnx模型转换示例
\ No newline at end of file
+
+将caffe, pytorch, tf, tflite, paddle 模型转成onnx模型的示例。
+
+## caffe转onnx
+
+caffe转onnx请参考[caffe/caffe2onnx/README.md](caffe/caffe2onnx/README.md)
+
+## paddle转onnx
+
+paddle转onnx请参考[paddle/README.md](paddle/README.md)
+
+## tf以及tflite转onnx
+
+tf以及tflite请参考[tf_tflite/README.md](tf_tflite/README.md)
+
+## torch转onnx
+
+torch转onnx请参考[torch/README.md](torch/README.md)
+

caffe/caffe2onnx/.gitignore

+### https://raw.github.com/github/gitignore/2f75277037d172200d4a37621c1b9c3b9901dbd8/Global/JetBrains.gitignore
+
+# Covers JetBrains IDEs: IntelliJ, RubyMine, PhpStorm, AppCode, PyCharm, CLion, Android Studio, WebStorm and Rider
+# Reference: https://intellij-support.jetbrains.com/hc/en-us/articles/206544839
+
+# User-specific stuff
+.idea/**/workspace.xml
+.idea/**/tasks.xml
+.idea/**/usage.statistics.xml
+.idea/**/dictionaries
+.idea/**/shelf
+
+# Generated files
+.idea/**/contentModel.xml
+
+# Sensitive or high-churn files
+.idea/**/dataSources/
+.idea/**/dataSources.ids
+.idea/**/dataSources.local.xml
+.idea/**/sqlDataSources.xml
+.idea/**/dynamic.xml
+.idea/**/uiDesigner.xml
+.idea/**/dbnavigator.xml
+
+# Gradle
+.idea/**/gradle.xml
+.idea/**/libraries
+
+# Gradle and Maven with auto-import
+# When using Gradle or Maven with auto-import, you should exclude module files,
+# since they will be recreated, and may cause churn.  Uncomment if using
+# auto-import.
+# .idea/artifacts
+# .idea/compiler.xml
+# .idea/jarRepositories.xml
+# .idea/modules.xml
+# .idea/*.iml
+# .idea/modules
+# *.iml
+# *.ipr
+
+# CMake
+cmake-build-*/
+
+# Mongo Explorer plugin
+.idea/**/mongoSettings.xml
+
+# File-based project format
+*.iws
+
+# IntelliJ
+out/
+
+# mpeltonen/sbt-idea plugin
+.idea_modules/
+
+# JIRA plugin
+atlassian-ide-plugin.xml
+
+# Cursive Clojure plugin
+.idea/replstate.xml
+
+# Crashlytics plugin (for Android Studio and IntelliJ)
+com_crashlytics_export_strings.xml
+crashlytics.properties
+crashlytics-build.properties
+fabric.properties
+
+# Editor-based Rest Client
+.idea/httpRequests
+
+# Android studio 3.1+ serialized cache file
+.idea/caches/build_file_checksums.ser
+
+
+### https://raw.github.com/github/gitignore/2f75277037d172200d4a37621c1b9c3b9901dbd8/Python.gitignore
+
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+pip-wheel-metadata/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+
+### https://raw.github.com/github/gitignore/2f75277037d172200d4a37621c1b9c3b9901dbd8/Global/macOS.gitignore
+
+# General
+.DS_Store
+.AppleDouble
+.LSOverride
+
+# Icon must end with two \r
+Icon
+
+# Thumbnails
+._*
+
+# Files that might appear in the root of a volume
+.DocumentRevisions-V100
+.fseventsd
+.Spotlight-V100
+.TemporaryItems
+.Trashes
+.VolumeIcon.icns
+.com.apple.timemachine.donotpresent
+
+# Directories potentially created on remote AFP share
+.AppleDB
+.AppleDesktop
+Network Trash Folder
+Temporary Items
+.apdisk
+
+

caffe/caffe2onnx/README.md

+# Caffe 模型转换为 ONNX 模型
+
+## 1. 环境搭建
+
+- onnx(version == 1.6.0)
+```shell script
+pip3 install onnx==1.6.0
+```
+
+- numpy(version >= 1.17.0)
+```shell script
+pip3 install numpy
+```
+
+## 2. caffe2onnx 工具使用
+- 进入工具目录
+``` shell script
+cd <tnn_root_path>/tools/caffe2onnx/
+```
+
+- 模型下载
+示例caffe模型[下载地址](https://github.com/GammaLab-HPC/dawnbench_inference_imagenet/tree/master/model),prototxt和caffemodel均需要下载。
+
+- 使用转换工具
+```shell script
+python3 convert2onnx.py ./resnet50.prototxt ./resnet50.caffemodel -o ./test.onnx
+```
+- 转换工具脚本参数说明
+```text
+usage: caffe2onnx.py [-h] [-o ONNX_FILE] proto_file caffe_model_file
+
+convert caffe model to onnx
+
+positional arguments:
+  proto_file        the path for prototxt file, the file name must end with .prototxt
+  caffe_model_file  the path for caffe model file, the file name must end with .caffemodel!
+
+options:
+  -h, --help        show this help message and exit
+  -o ONNX_FILE      the path for generate onnx file
+```
\ No newline at end of file

caffe/caffe2onnx/caffe2onnx.py

+from src.load_save_model import LoadCaffeModel, SaveOnnxModel
+from src.caffe2onnx import Caffe2Onnx
+from src.args_parser import parse_args
+from src.utils import is_ssd_model
+
+def main(args):
+    caffe_graph_path = args.proto_file
+    caffe_params_path = args.caffe_model_file
+
+    pos_s = caffe_graph_path.rfind("/")
+    if  pos_s == -1:
+        pos_s = 0
+
+    pos_dot = caffe_graph_path.rfind(".")
+    onnx_name = caffe_graph_path[pos_s+1:pos_dot]
+    save_path = caffe_graph_path[0:pos_dot] + '.onnx'
+    if args.onnx_file is not None:
+        save_path = args.onnx_file
+
+    graph, params = LoadCaffeModel(caffe_graph_path,caffe_params_path)
+    print('2. 开始进行模型转换')
+    c2o = Caffe2Onnx(graph, params, onnx_name)
+    print('3. 创建 onnx 模型')
+    onnx_model = c2o.createOnnxModel()
+    print('4. 保存 onnx 模型')
+    # is_ssd = is_ssd_model(caffe_graph_path)
+    # if is_ssd:
+    SaveOnnxModel(onnx_model, save_path, need_polish=False)
+    # else:
+    #     SaveOnnxModel(onnx_model, save_path, need_polish=True)
+
+
+if __name__ == '__main__':
+    args = parse_args()
+    main(args)

caffe/caffe2onnx/doc/onnx-version.md

+# onnx version
+
+| onnx                  | 1.2.2                              | 1.6.0                                            | compatible   |
+|-----------------------|------------------------------------|--------------------------------------------------|--------------|
+| AveragePool           | -                                  | attributes(ceil\_mode)                           | yes          |
+| BatchNormalization    | spatial                            | spatial(delete) (not use)                        | yes          |
+| Clip                  | attributes(min, max)               | inputs(min, max)                                 | yes(support) |
+| Concat                | -                                  | -                                                | yes          |
+| Conv                  | -                                  | -                                                | yes          |
+| ConvTranspose         | -                                  | -                                                | yes          |
+| DepthToSpace          | attributes(blocksize)              | attributes(blocksize,mode)                       | yes(support) |
+| Div                   | -                                  | -                                                | yes          |
+| Exp                   | -                                  | -                                                | yes          |
+| Expand                | not support                        | support                                          | yes          |
+| Gemm                  | -                                  | -                                                | yes          |
+| GlobalAveragePool     | -                                  | -                                                | yes          |
+| GlobalMaxPool         | -                                  | -                                                | yes          |
+| InstanceNormalization | -                                  | -                                                | yes          |
+| LeakyRelu             | -                                  | -                                                | yes          |
+| MaxPool               | -                                  | attributes(ceil\_mode, dilations,storage\_order) | ?            |
+| Mul                   | -                                  | -                                                | yes          |
+| PRelu                 | -                                  | -                                                | yes          |
+| Pad                   | attributes(pads, value)            | inputs(pads, constant_value)                     | yes(support) |
+| ReduceL2              | -                                  | -                                                | yes          |
+| ReduceMean            | -                                  | -                                                | yes          |
+| ReduceSum             | -                                  | -                                                | yes          |
+| Relu                  | -                                  | -                                                | yes          |
+| Reshape               | -                                  | -                                                | yes          |
+| Slice                 | attributes(starts,ends,axes,steps) | inputs(starts,ends,axes,steps)                   | yes(support) |
+| Softmax               | -                                  | -                                                | yes          |
+| Softplus              | -                                  | -                                                | yes          |
+| Split                 | -                                  | -                                                | yes          |
+| Sub                   | -                                  | -                                                | yes          |
+| Tanh                  | -                                  | -                                                | yes          |
+| Tile                  | -                                  | -                                                | yes          |
+| Transpose             | -                                  | -                                                | yes          |
+| Upsample              | Upsample                           | deprecated(弃用了), 使用 Resize 替代             | yes          |
+    

caffe/caffe2onnx/src/OPs/Add.py

+# Tencent is pleased to support the open source community by making TNN available.
+#
+# Copyright (C) 2020 THL A29 Limited, a Tencent company. All rights reserved.
+#
+# Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
+# in compliance with the License. You may obtain a copy of the License at
+#
+# https://opensource.org/licenses/BSD-3-Clause
+#
+# 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.
+
+import src.c2oObject as Node
+
+
+def get_add_output_shape(input_shape):
+    output_shape = input_shape[0]
+    return [output_shape]
+
+
+def create_add_node(layer, node_name, input_name, output_name, input_shape):
+    output_shape = get_add_output_shape(input_shape)
+
+    node = Node.c2oNode(layer, node_name, 'Add', input_name, output_name, input_shape, output_shape)
+
+    return node

caffe/caffe2onnx/src/OPs/Axpy.py

+# Tencent is pleased to support the open source community by making TNN available.
+#
+# Copyright (C) 2020 THL A29 Limited, a Tencent company. All rights reserved.
+#
+# Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
+# in compliance with the License. You may obtain a copy of the License at
+#
+# https://opensource.org/licenses/BSD-3-Clause
+#
+# 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.
+
+import src.c2oObject as Node
+from typing import *
+import copy
+
+
+def need_add_reshape(input_shape: List[List]) -> bool:
+    return len(input_shape[0]) != len(input_shape[1])
+
+
+def get_param_shape(input_shape: List[List]) -> List:
+    input = input_shape[0]
+    scale = copy.deepcopy(input_shape[1])
+    if len(input) > len(scale):
+        for i in range(len(input) - len(scale)):
+            scale.append(1)
+    return scale
+
+def broadcast_scale(input_shape: List[List]) -> List[List]:
+    input = input_shape[0]
+    scale = input_shape[1]
+    if len(input) > len(scale):
+        for i in range(len(input) - len(scale)):
+            scale.append(1)
+        broadcast_shape = [input, scale]
+    elif len(input) < len(scale):
+        print("the scale should be less than input")
+        exit(-1)
+    else:
+        broadcast_shape = [input, scale]
+    return broadcast_shape
+
+
+def get_mul_output_shape(input_shape: List[List]) -> List[List]:
+    output_shape = input_shape[1]
+    return [output_shape]
+
+
+def create_axpy_mul_node(layer, node_name, input_name, output_name, input_shape):
+
+    new_node_name = node_name + "_middle"
+    output_shape = get_mul_output_shape(input_shape)
+    new_input_name = [input_name[0], input_name[1]]
+    new_output_name = [output_name[0] + "_mul"]
+    new_input_shape = [input_shape[0], input_shape[1]]
+
+    node = Node.c2oNode(layer, new_node_name, 'Mul', new_input_name, new_output_name, new_input_shape, output_shape)
+
+    return node
+
+def get_add_output_shape(input_shape):
+
+    output_shape = input_shape[1]
+
+    return [output_shape]
+
+def create_axpy_add_node(layer, node_name, input_name, output_name, input_shape):
+
+    output_shape = get_add_output_shape(input_shape)
+    new_input_name = [node_name + "_mul", input_name[2]]
+    new_input_shape = [input_shape[1], input_shape[2]]
+    node = Node.c2oNode(layer, node_name, "Add", new_input_name, output_name, input_shape, output_shape)
+
+    return node

caffe/caffe2onnx/src/OPs/BatchNorm.py

+import src.c2oObject as Node
+
+##-----------------------------BatchNormalization层 = BatchNorm + Scale-------------------------------------##
+#获取超参数
+def getBNAttri(layer):
+    #超参数字典
+    eps = layer.batch_norm_param.eps
+    momentum = layer.batch_norm_param.moving_average_fraction
+    dict = {"epsilon": eps,  # 滑动系数
+            "momentum": momentum
+            }
+    return dict
+#计算输出维度
+def getBNOutShape(input_shape):
+    output_shape = input_shape
+    return output_shape
+#构建节点
+def createBN(layer, nodename, inname, outname, input_shape):
+    dict = getBNAttri(layer)
+    #计算output_shape,输出维度等于输入维度
+    output_shape = getBNOutShape(input_shape)
+
+    #构建node
+    node = Node.c2oNode(layer, nodename, "BatchNormalization", inname, outname, input_shape, output_shape,dict)
+    return node
\ No newline at end of file

caffe/caffe2onnx/src/OPs/Clip.py

+# Tencent is pleased to support the open source community by making TNN available.
+#
+# Copyright (C) 2020 THL A29 Limited, a Tencent company. All rights reserved.
+#
+# Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
+# in compliance with the License. You may obtain a copy of the License at
+#
+# https://opensource.org/licenses/BSD-3-Clause
+#
+# 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.
+
+import src.c2oObject as Node
+
+def get_attribute(layer):
+    attributes = {}
+    max_attribute = 0
+    min_attribute = 0
+    if layer.type == 'ReLU6':
+        max_attribute = 6.0
+        min_attribute = 0
+
+    attribute = {
+        'max': max_attribute,
+        'min': min_attribute
+    }
+    return attributes
+
+
+def get_clip_output_shape(input_shape):
+    output_shape = input_shape
+    return output_shape
+
+
+def create_clip_node(layer, node_name, input_name, output_name, input_shape):
+    # onnx 1.6.0 don't use
+    # attributes = get_attribute(layer)
+    output_shape = get_clip_output_shape(input_shape)
+    node = Node.c2oNode(layer, node_name, 'Clip', input_name, output_name, input_shape, output_shape)
+    return node

caffe/caffe2onnx/src/OPs/Concat.py

+import src.c2oObject as Node
+from typing import List
+import copy
+
+
+def get_concat_attributes(layer):
+    ##轴
+    axis = layer.concat_param.axis
+    attributes = {"axis": axis}
+    return attributes
+
+
+# 计算输出维度
+def get_concat_outshape(layer, input_shape: List) -> List:
+    bottom = input_shape[0]
+    axis = layer.concat_param.axis
+
+    output_shape = copy.deepcopy(bottom)
+
+    assert (axis < len(bottom))
+
+    for i in range(1, len(input_shape)):
+        output_shape[axis] = output_shape[axis] + input_shape[i][axis]
+    return [output_shape]
+    #
+    # if len(bottom) == 2:
+    #     n, c = bottom[0], 0
+    #     for i in range(len(input_shape)):
+    #         c = c + input_shape[i][1]
+    #     output_shape = [[n, c]]
+    #     return output_shape
+    #
+    # elif len(bottom) == 3:
+    #     n, c = bottom[0], 0
+    #     for i in range(len(input_shape)):
+    #         c = c + input_shape[i][1]
+    #     output_shape = [[n, c]]
+    #     return output_shape
+    #
+    # elif len(bottom) == 4:
+    #     n, c, w, h = input_shape[0][0], 0, input_shape[0][2], input_shape[0][3]
+    #     for i in range(len(input_shape)):
+    #         c = c + input_shape[i][1]
+    #     output_shape = [[n, c, w, h]]
+    #     return output_shape
+
+
+# 构建节点
+def createConcat(layer, nodename, inname, outname, input_shape):
+    attributes = get_concat_attributes(layer)
+
+    output_shape = get_concat_outshape(layer, input_shape)
+
+    node = Node.c2oNode(layer, nodename, "Concat", inname, outname, input_shape, output_shape, attributes)
+
+    return node

caffe/caffe2onnx/src/OPs/Conv.py

+import numpy as np
+import src.c2oObject as Node
+import math
+
+
+# 获取超参数
+def getConvAttri(layer, input_shape):
+    # 膨胀系数 dilations
+    dilations = [1, 1]
+    if layer.convolution_param.dilation != []:
+        dilation = layer.convolution_param.dilation[0]
+        dilations = [dilation, dilation]
+    ##填充pads
+    pads = [0, 0, 0, 0]  # 默认为0
+    if layer.convolution_param.pad != []:  # 若存在pad,则根据pad赋值
+        pads = np.array([layer.convolution_param.pad] * 4).reshape(1, -1)[0].tolist()
+    elif layer.convolution_param.pad_h != 0 or layer.convolution_param.pad_w != 0:  # 若存在pad_w,pad_h则根据其赋值
+        pads = [layer.convolution_param.pad_h, layer.convolution_param.pad_w, layer.convolution_param.pad_h,
+                layer.convolution_param.pad_w]
+    ##步长strides
+    strides = [1, 1]  # 默认为1
+    if layer.convolution_param.stride != []:
+        strides = np.array([layer.convolution_param.stride] * 2).reshape(1, -1)[0].tolist()
+
+    elif layer.convolution_param.stride_h != 0 and layer.convolution_param.stride_w != 0:
+        strides = [layer.convolution_param.stride_h, layer.convolution_param.stride_w]
+
+    ##卷积核尺寸kernel_shape
+    kernel_shape = np.array([layer.convolution_param.kernel_size] * 2).reshape(1, -1)[0].tolist()
+    if layer.convolution_param.kernel_size == []:
+        kernel_shape = [layer.convolution_param.kernel_h, layer.convolution_param.kernel_w]
+    ##分组group
+    group = 1
+    if layer.type == "ConvolutionDepthwise":
+        group = input_shape[0][1]
+    else:
+        group = layer.convolution_param.group
+
+    # 超参数字典
+    dict = {
+        #"auto_pad":"NOTSET",
+        "dilations": dilations,
+        "group": group,
+        "kernel_shape": kernel_shape,
+        "pads": pads,
+        "strides": strides
+    }
+    return dict
+
+
+# 计算输出维度
+def getConvOutShape(input_shape, layer, dict):
+    dilations = dict["dilations"]
+    kernel_shape = dict["kernel_shape"]
+    pads = dict["pads"]
+    strides = dict["strides"]
+    ##卷积核数量kernel_num
+    kernel_num = layer.convolution_param.num_output
+    # reference the caffe source code
+    kernel_extent_h = dilations[0] * (kernel_shape[0] - 1) + 1
+    output_shape_h = math.floor((input_shape[0][2] + 2 * pads[0] - kernel_extent_h) / strides[0]) + 1
+
+    kernel_extent_w = dilations[1] * (kernel_shape[1] - 1) + 1
+    output_shape_w = math.floor((input_shape[0][3] + 2 * pads[1] - kernel_extent_w) / strides[1]) + 1
+
+    output_shape = [[input_shape[0][0], kernel_num, output_shape_h, output_shape_w]]
+    return output_shape
+
+
+# 构建节点
+def createConv(layer, node_name, input_name, output_name, input_shape):
+    attributes = getConvAttri(layer, input_shape)
+    output_shape = getConvOutShape(input_shape, layer, attributes)
+    # 构建node
+    node = Node.c2oNode(layer, node_name, "Conv", input_name, output_name, input_shape, output_shape, attributes)
+    return node

caffe/caffe2onnx/src/OPs/ConvTranspose.py

+import numpy as np
+import src.c2oObject as Node
+##---------------------------------------------------ConvTranspose层-------------------------------------------------------##
+#获取超参数
+def getConvTransposeAttri(layer):
+    ##膨胀系数dilations
+    dilations = [1, 1]
+    if layer.convolution_param.dilation != []:
+        dilation = layer.convolution_param.dilation[0]
+        dilations = [dilation, dilation]
+    ##填充pads
+    pads = [0, 0, 0, 0]  # 默认为0
+    if layer.convolution_param.pad != []:  # 若存在pad,则根据pad赋值
+        pads = np.array([layer.convolution_param.pad] * 4).reshape(1, -1)[0].tolist()
+    elif layer.convolution_param.pad_h != 0 or layer.convolution_param.pad_w != 0:  # 若存在pad_w,pad_h则根据其赋值
+        pads = [layer.convolution_param.pad_h, layer.convolution_param.pad_w, layer.convolution_param.pad_h,
+                layer.convolution_param.pad_w]
+    ##步长strides
+    strides = [1, 1]  # 默认为1
+    if layer.convolution_param.stride != []:
+        strides = np.array([layer.convolution_param.stride] * 2).reshape(1, -1)[0].tolist()
+    elif layer.convolution_param.stride_h != 0 and layer.convolution_param.stride_w != 0:
+        strides = [layer.convolution_param.stride_h, layer.convolution_param.stride_w]
+
+    ##卷积核尺寸kernel_shape
+    kernel_shape = np.array([layer.convolution_param.kernel_size] * 2).reshape(1, -1)[0].tolist()
+    if layer.convolution_param.kernel_size == []:
+        kernel_shape = [layer.convolution_param.kernel_h, layer.convolution_param.kernel_w]
+    ##分组group
+    group = layer.convolution_param.group
+
+
+    # 超参数字典
+    dict = {  # "auto_pad":"NOTSET",
+        "dilations": dilations,
+        "group": group,
+        "kernel_shape": kernel_shape,
+        "pads": pads,
+        "strides": strides
+    }
+    return dict
+#计算输出维度
+def getConvTransposeOutShape(input_shape, layer,dict):
+    dilations = dict["dilations"]
+    kernel_shape = dict["kernel_shape"]
+    pads = dict["pads"]
+    strides = dict["strides"]
+    ##卷积核数量kernel_num
+    kernel_num = layer.convolution_param.num_output
+
+    def get_output_shape(i, k, p, s):
+        return (i-1)*s + k -2*p
+
+    h = get_output_shape(input_shape[0][2], kernel_shape[0], pads[0], strides[0])
+    w = get_output_shape(input_shape[0][3], kernel_shape[1], pads[1], strides[1])
+
+    output_shape = [[input_shape[0][0], kernel_num, h, w]]
+
+    return output_shape
+#构建节点
+def createConvTranspose(layer, nodename, inname, outname, input_shape):
+    dict = getConvTransposeAttri(layer)
+    output_shape = getConvTransposeOutShape(input_shape, layer, dict)
+    #构建node
+    node = Node.c2oNode(layer, nodename, "ConvTranspose", inname, outname, input_shape, output_shape, dict)
+    return node

caffe/caffe2onnx/src/OPs/Crop.py

+# Tencent is pleased to support the open source community by making TNN available.
+#
+# Copyright (C) 2020 THL A29 Limited, a Tencent company. All rights reserved.
+#
+# Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
+# in compliance with the License. You may obtain a copy of the License at
+#
+# https://opensource.org/licenses/BSD-3-Clause
+#
+# 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.
+
+import src.c2oObject as Node
+import numpy as np
+
+def get_crop_param(layer, input_shape):
+    axis: int = layer.crop_param.axis
+    crop_offset = layer.crop_param.offset
+
+    if not crop_offset:
+        offset_0 = 0
+    else:
+        offset_0 = crop_offset[0]
+
+    offset = []
+    starts = []
+    axes = []
+    ends = []
+
+    for i in range(len(input_shape[0])):
+        if i < axis:
+            start = 0
+            end = input_shape[1][i]
+        else:
+            if (i - axis) >= len(crop_offset):
+                offset.append(offset_0)
+            else:
+                offset.append(crop_offset[i - axis])
+
+            start = offset[i - axis]
+            end = start + input_shape[1][i]
+
+        if input_shape[0][i] != input_shape[1][i]:
+            axes.append(i)
+            starts.append(start)
+            ends.append(end)
+
+    return starts, ends, axes
+
+def get_crop_output_shape(layer, input_shape):  
+    return [input_shape[1]]
+
+
+def create_crop_node(layer, node_name, input_name, output_name, input_shape):
+    output_shape = get_crop_output_shape(layer, input_shape)
+    node = Node.c2oNode(layer, node_name, "Slice", input_name, output_name, input_shape, output_shape)
+    return node

caffe/caffe2onnx/src/OPs/DetectionOutput.py

+# Tencent is pleased to support the open source community by making TNN available.
+#
+# Copyright (C) 2020 THL A29 Limited, a Tencent company. All rights reserved.
+#
+# Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
+# in compliance with the License. You may obtain a copy of the License at
+#
+# https://opensource.org/licenses/BSD-3-Clause
+#
+# 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.
+
+import onnx
+from typing import *
+from onnx import helper
+from typing import *
+import ctypes
+import src.c2oObject as Node
+
+
+def create_attribuates(layer) -> Dict:
+    detection_output_param = layer.detection_output_param
+    num_classes  = detection_output_param.num_classes
+    share_location        = 1 if detection_output_param.share_location else 0
+    background_label_id    = detection_output_param.background_label_id
+    # NonMaximumSuppressionParameter
+    nms_threshold = detection_output_param.nms_param.nms_threshold
+    top_k = detection_output_param.nms_param.top_k
+    eta = detection_output_param.nms_param.eta
+
+    code_type              = detection_output_param.code_type
+    variance_encoded_in_target = 1 if detection_output_param.variance_encoded_in_target else 0
+    keep_top_k  = detection_output_param.keep_top_k
+    confidence_threshold = detection_output_param.confidence_threshold
+    visualize = 1 if detection_output_param.visualize else 0
+    visualize_threshold = detection_output_param.visualize_threshold
+    save_file = detection_output_param.save_file
+
+
+
+    # TODO: SaveOutputParameter
+    # save_output_param = detection_output_param.save_output_param
+    # output_directory: str = save_output_param.output_directory
+    # output_name_prefix: str = save_output_param.output_name_prefix
+    # output_format: str = save_output_param.output_format
+    # label_map_file: str = save_output_param.label_map_file
+    # name_size_file: str = save_output_param.name_size_file
+    # num_test_image: int = save_output_param.num_test_image
+
+
+
+    attributes = {
+        'num_classes'            : num_classes,
+        'share_location'       : share_location,
+        'background_label_id'  : background_label_id,
+        'nms_threshold'        : nms_threshold,
+        'top_k'                : top_k,
+        'eta'                  : eta,
+        'code_type'            : code_type,
+        'variance_encoded_in_target' : variance_encoded_in_target,
+        'keep_top_k'           : keep_top_k,
+        'confidence_threshold' : confidence_threshold,
+        'visualize'            : visualize,
+        'visualize_threshold'  : visualize_threshold,
+        'save_file'            : save_file
+        }
+    return attributes
+
+
+def create_detection_output(layer,
+                            node_name: str,
+                            inputs_name: List[str],
+                            outputs_name: List[str],
+                            inputs_shape: List, ) -> onnx.NodeProto:
+
+    attributes = create_attribuates(layer)
+
+    outputs_shape = [[1, 1, 1, 7]]
+
+    node = Node.c2oNode(layer, node_name, "DetectionOutput",
+                        inputs_name, outputs_name,
+                        inputs_shape, outputs_shape,
+                        attributes)
+    return node

caffe/caffe2onnx/src/OPs/Dropout.py

+import src.c2oObject as Node
+##----------------------------------------------------Dropout层-------------------------------------------------------##
+#获取超参数
+def getDropoutAttri(layer):
+    ##drop 比率
+    ratio = layer.dropout_param.dropout_ratio
+    #前向不需要dropout,ratio设置为0后，后续可以onnx工具优化掉
+    ratio = 0.0
+
+    # 超参数字典
+    dict = {"ratio":ratio}
+    return dict
+def getDropoutOutShape(input_shape):
+    # 计算输出维度output_shape
+    output_shape = input_shape  # 与输入维度一样
+    return output_shape
+#构建节点
+def createDropout(layer, nodename, inname, outname, input_shape):
+    dict = getDropoutAttri(layer)
+    output_shape = getDropoutOutShape(input_shape)
+    # 构建node
+    node = Node.c2oNode(layer, nodename, "Dropout", inname, outname, input_shape, output_shape, dict=dict)
+    return node

caffe/caffe2onnx/src/OPs/Eltwise.py

+import src.c2oObject as Node
+##-------------------------------------------------eltwise层----------------------------------------------------------##
+def createEltwise(layer, nodename, inname, outname, input_shape):
+    #判断算子类型
+    if layer.eltwise_param.operation == 0:
+        node = __createMul(layer, nodename, inname, outname, input_shape)#按元素相乘
+
+    elif layer.eltwise_param.operation == 1:
+        node = __createAdd(layer, nodename, inname, outname, input_shape)#按元素相加
+
+    elif layer.eltwise_param.operation == 2:
+        node = __createMax(layer, nodename, inname, outname, input_shape)#按元素求最大值
+
+    return node
+
+
+
+##----------------------------------------------Mul层,对应Prod-----------------------------------------------##
+def __createMul(layer, nodename, inname, outname, input_shape):
+    output_shape = input_shape[0]
+    node = Node.c2oNode(layer, nodename, "Mul", inname, outname, input_shape, output_shape)
+    return node
+
+##---------------------Add层,可能是两个中间层输出相加，也可能是一个输出加一个bias这种------------------------##
+def __createAdd(layer, nodename, inname, outname, input_shape):
+    output_shape = [input_shape[0]]
+    node = Node.c2oNode(layer, nodename, "Add", inname, outname, input_shape, output_shape)
+    return node
+##----------------------------------------------Max层-------------------------------------------------------------##
+def __createMax(layer, nodename, inname, outname, input_shape):
+    output_shape = input_shape
+    node = Node.c2oNode(layer, nodename, "Max", inname, outname, input_shape, output_shape)
+    return node

caffe/caffe2onnx/src/OPs/Flatten.py

+import src.c2oObject as Node
+from typing import List, Dict
+import onnx
+
+
+def get_attributes(layer) -> Dict:
+    axis = layer.flatten_param.axis
+    end_axis = layer.flatten_param.end_axis
+    if end_axis != -1:
+        print("not support end_axis param!")
+        exit(-1)
+    attributes = {
+        "axis": axis
+    }
+    return attributes
+
+
+def get_flatten_output_shape(input_shape: List,
+                             attributes: Dict) -> List:
+    shape = input_shape[0]
+    input_prod = 1
+    axis = attributes.get("axis")
+    for i in range(axis, len(shape)):
+        input_prod = input_prod * shape[i]
+
+    output_shape = [shape[0:axis]+ [input_prod]]
+    return output_shape
+
+def create_flatten_node(layer, node_name : str,
+                        input_names: List,
+                        output_name: List,
+                        input_shape: List) -> onnx.NodeProto:
+    attributes = get_attributes(layer)
+
+    output_shape = get_flatten_output_shape(input_shape, attributes)
+
+    node = Node.c2oNode(layer, node_name, "Flatten", input_names,
+                        output_name, input_shape, output_shape, attributes)
+    return node