[Backend] add sophgo backend (#1015)

* Add Sophgo Device

add sophgo backend in fastdeploy

add resnet50, yolov5s, liteseg examples.

* replace sophgo lib with download links; fix model.cc bug

* modify CodeStyle

* remove unuseful files;change the names of sophgo device and sophgo
backend

* sophgo support python and add python examples

* remove unuseful rows in cmake according pr

Co-authored-by: Zilong Xing <zilong.xing@sophgo.com>

examples/vision/segmentation/paddleseg/sophgo/README.md

@@ -0,0 +1,89 @@
+# PaddleSeg C++部署示例
+
+## 支持模型列表
+
+- PP-LiteSeg部署模型实现来自[PaddleSeg PP-LiteSeg系列模型](https://github.com/PaddlePaddle/PaddleSeg/blob/release/2.6/configs/pp_liteseg/README.md)
+
+## 准备PP-LiteSeg部署模型以及转换模型
+
+SOPHGO-TPU部署模型前需要将Paddle模型转换成bmodel模型，具体步骤如下:
+- 下载Paddle模型[PP-LiteSeg-B(STDC2)-cityscapes-without-argmax](https://bj.bcebos.com/paddlehub/fastdeploy/PP_LiteSeg_B_STDC2_cityscapes_without_argmax_infer.tgz)
+- Pddle模型转换为ONNX模型，请参考[Paddle2ONNX](https://github.com/PaddlePaddle/Paddle2ONNX)
+- ONNX模型转换bmodel模型的过程，请参考[TPU-MLIR](https://github.com/sophgo/tpu-mlir)
+
+## 模型转换example
+
+下面以[PP-LiteSeg-B(STDC2)-cityscapes-without-argmax](https://bj.bcebos.com/paddlehub/fastdeploy/PP_LiteSeg_B_STDC2_cityscapes_without_argmax_infer.tgz)为例子,教大家如何转换Paddle模型到SOPHGO-TPU模型
+
+### 下载PP-LiteSeg-B(STDC2)-cityscapes-without-argmax模型,并转换为ONNX模型
+```shell
+https://bj.bcebos.com/paddlehub/fastdeploy/PP_LiteSeg_B_STDC2_cityscapes_without_argmax_infer.tgz
+tar xvf PP_LiteSeg_B_STDC2_cityscapes_without_argmax_infer.tgz
+
+# 修改PP_LiteSeg_B_STDC2_cityscapes_without_argmax_infer模型的输入shape，由动态输入变成固定输入
+python paddle_infer_shape.py --model_dir PP_LiteSeg_B_STDC2_cityscapes_without_argmax_infer \
+                             --model_filename model.pdmodel \
+                             --params_filename model.pdiparams \
+                             --save_dir pp_liteseg_fix \
+                             --input_shape_dict="{'x':[1,3,512,512]}"
+
+#将固定输入的Paddle模型转换成ONNX模型
+paddle2onnx --model_dir pp_liteseg_fix \
+            --model_filename model.pdmodel \
+            --params_filename model.pdiparams \
+            --save_file pp_liteseg.onnx \
+            --enable_dev_version True
+```
+
+### 导出bmodel模型
+
+以转换BM1684x的bmodel模型为例子，我们需要下载[TPU-MLIR](https://github.com/sophgo/tpu-mlir)工程，安装过程具体参见[TPU-MLIR文档](https://github.com/sophgo/tpu-mlir/blob/master/README.md)。
+### 1.	安装
+``` shell
+docker pull sophgo/tpuc_dev:latest
+
+# myname1234是一个示例，也可以设置其他名字
+docker run --privileged --name myname1234 -v $PWD:/workspace -it sophgo/tpuc_dev:latest
+
+source ./envsetup.sh
+./build.sh
+```
+
+### 2.	ONNX模型转换为bmodel模型
+``` shell
+mkdir pp_liteseg && cd pp_liteseg
+
+#在该文件中放入测试图片，同时将上一步转换的pp_liteseg.onnx放入该文件夹中
+cp -rf ${REGRESSION_PATH}/dataset/COCO2017 .
+cp -rf ${REGRESSION_PATH}/image .
+#放入onnx模型文件pp_liteseg.onnx
+
+mkdir workspace && cd workspace
+
+#将ONNX模型转换为mlir模型，其中参数--output_names可以通过NETRON查看
+model_transform.py \
+    --model_name pp_liteseg \
+    --model_def ../pp_liteseg.onnx \
+    --input_shapes [[1,3,512,512]] \
+    --mean 0.0,0.0,0.0 \
+    --scale 0.0039216,0.0039216,0.0039216 \
+    --keep_aspect_ratio \
+    --pixel_format rgb \
+    --output_names bilinear_interp_v2_6.tmp_0 \
+    --test_input ../image/dog.jpg \
+    --test_result pp_liteseg_top_outputs.npz \
+    --mlir pp_liteseg.mlir
+
+#将mlir模型转换为BM1684x的F32 bmodel模型
+model_deploy.py \
+  --mlir pp_liteseg.mlir \
+  --quantize F32 \
+  --chip bm1684x \
+  --test_input pp_liteseg_in_f32.npz \
+  --test_reference pp_liteseg_top_outputs.npz \
+  --model pp_liteseg_1684x_f32.bmodel
+```
+最终获得可以在BM1684x上能够运行的bmodel模型pp_liteseg_1684x_f32.bmodel。如果需要进一步对模型进行加速，可以将ONNX模型转换为INT8 bmodel，具体步骤参见[TPU-MLIR文档](https://github.com/sophgo/tpu-mlir/blob/master/README.md)。
+
+## 其他链接
+- [Cpp部署](./cpp)

examples/vision/segmentation/paddleseg/sophgo/cpp/CMakeLists.txt

@@ -0,0 +1,17 @@
+PROJECT(infer_demo C CXX)
+CMAKE_MINIMUM_REQUIRED (VERSION 3.10)
+# 指定下载解压后的fastdeploy库路径
+option(FASTDEPLOY_INSTALL_DIR "Path of downloaded fastdeploy sdk.")
+
+set(ENABLE_LITE_BACKEND OFF)
+#set(FDLIB ${FASTDEPLOY_INSTALL_DIR})
+
+include(${FASTDEPLOY_INSTALL_DIR}/FastDeploy.cmake)
+
+# 添加FastDeploy依赖头文件
+include_directories(${FASTDEPLOY_INCS})
+include_directories(${FastDeploy_INCLUDE_DIRS})
+
+add_executable(infer_demo ${PROJECT_SOURCE_DIR}/infer.cc)
+# 添加FastDeploy库依赖
+target_link_libraries(infer_demo ${FASTDEPLOY_LIBS})

examples/vision/segmentation/paddleseg/sophgo/cpp/README.md

@@ -0,0 +1,56 @@
+# PaddleSeg C++部署示例
+
+本目录下提供`infer.cc`快速完成pp_liteseg模型在SOPHGO BM1684x板子上加速部署的示例。
+
+在部署前，需确认以下两个步骤:
+
+1. 软硬件环境满足要求
+2. 根据开发环境，从头编译FastDeploy仓库
+
+以上步骤请参考[SOPHGO部署库编译](../../../../../../docs/cn/build_and_install/sophgo.md)实现
+
+## 生成基本目录文件
+
+该例程由以下几个部分组成
+```text
+.
+├── CMakeLists.txt
+├── build  # 编译文件夹
+├── image  # 存放图片的文件夹
+├── infer.cc
+└── model  # 存放模型文件的文件夹
+```
+
+## 编译
+
+### 编译并拷贝SDK到thirdpartys文件夹
+
+请参考[SOPHGO部署库编译](../../../../../../docs/cn/build_and_install/sophgo.md)仓库编译SDK，编译完成后，将在build目录下生成fastdeploy-0.0.3目录.
+
+### 拷贝模型文件，以及配置文件至model文件夹
+将Paddle模型转换为SOPHGO bmodel模型，转换步骤参考[文档](../README.md)  
+将转换后的SOPHGO bmodel模型文件拷贝至model中
+
+### 准备测试图片至image文件夹
+```bash
+wget https://paddleseg.bj.bcebos.com/dygraph/demo/cityscapes_demo.png
+cp cityscapes_demo.png ./images
+```
+
+### 编译example
+
+```bash
+cd build
+cmake .. -DFASTDEPLOY_INSTALL_DIR=${PWD}/fastdeploy-0.0.3
+make
+```
+
+## 运行例程
+
+```bash
+./infer_demo model images/cityscapes_demo.png
+```
+
+
+- [模型介绍](../../)
+- [模型转换](../)

examples/vision/segmentation/paddleseg/sophgo/cpp/infer.cc

@@ -0,0 +1,71 @@
+// 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 <iostream>
+#include <string>
+#include "fastdeploy/vision.h"
+
+void SophgoInfer(const std::string& model_dir, const std::string& image_file) {
+  std::string model_file = model_dir + "/pp_liteseg_1684x_f32.bmodel";
+  std::string params_file;
+  std::string config_file = model_dir + "/deploy.yaml";
+  auto option = fastdeploy::RuntimeOption();
+  option.UseSophgo();
+  auto model_format = fastdeploy::ModelFormat::SOPHGO;
+
+  auto model = fastdeploy::vision::segmentation::PaddleSegModel(
+      model_file, params_file, config_file, option, model_format);
+  if (!model.Initialized()) {
+    std::cerr << "Failed to initialize." << std::endl;
+    return;
+  }
+  //model.GetPreprocessor().DisableNormalizeAndPermute();
+
+  fastdeploy::TimeCounter tc;
+  tc.Start();
+  auto im_org = cv::imread(image_file);
+
+  //the input of bmodel should be fixed
+  int new_width = 512;
+  int new_height = 512;
+  cv::Mat im;
+  cv::resize(im_org, im, cv::Size(new_width, new_height), cv::INTER_LINEAR);
+
+  fastdeploy::vision::SegmentationResult res;
+  if (!model.Predict(&im, &res)) {
+    std::cerr << "Failed to predict." << std::endl;
+    return;
+  }
+  auto vis_im = fastdeploy::vision::VisSegmentation(im, res);
+  tc.End();
+  tc.PrintInfo("PPSeg in Sophgo");
+
+  cv::imwrite("infer_sophgo.jpg", vis_im);
+  std::cout
+      << "Visualized result saved in ./infer_sophgo.jpg"
+      << std::endl;
+}
+
+int main(int argc, char* argv[]) {
+  if (argc < 3) {
+    std::cout
+        << "Usage: infer_demo path/to/model_dir path/to/image run_option, "
+           "e.g ./infer_model ./bmodel ./test.jpeg"
+        << std::endl;
+    return -1;
+  }
+
+  SophgoInfer(argv[1], argv[2]);
+  return 0;
+}
+

examples/vision/segmentation/paddleseg/sophgo/python/README.md

@@ -0,0 +1,26 @@
+# PaddleSeg Python部署示例
+
+在部署前，需确认以下两个步骤
+
+- 1. 软硬件环境满足要求，参考[FastDeploy环境要求](../../../../../../docs/cn/build_and_install/sophgo.md)
+
+本目录下提供`infer.py`快速完成 pp_liteseg 在SOPHGO TPU上部署的示例。执行如下脚本即可完成
+
+```bash
+# 下载部署示例代码
+git clone https://github.com/PaddlePaddle/FastDeploy.git
+cd FastDeploy/examples/vision/segmentation/paddleseg/sophgo/python
+
+# 下载图片
+wget https://paddleseg.bj.bcebos.com/dygraph/demo/cityscapes_demo.png
+
+# 推理
+python3 infer.py --model_file ./bmodel/pp_liteseg_1684x_f32.bmodel --config_file ./bmodel/deploy.yaml --image cityscapes_demo.png
+
+# 运行完成后返回结果如下所示
+运行结果保存在sophgo_img.png中
+```
+
+## 其它文档
+- [pp_liteseg C++部署](../cpp)
+- [转换 pp_liteseg SOPHGO模型文档](../README.md)

examples/vision/segmentation/paddleseg/sophgo/python/infer.py

@@ -0,0 +1,45 @@
+import fastdeploy as fd
+import cv2
+import os
+
+
+def parse_arguments():
+    import argparse
+    import ast
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--model", required=True, help="Path of model.")
+    parser.add_argument(
+        "--config_file", required=True, help="Path of config file.")
+    parser.add_argument(
+        "--image", type=str, required=True, help="Path of test image file.")
+
+    return parser.parse_args()
+
+
+args = parse_arguments()
+
+# 配置runtime，加载模型
+runtime_option = fd.RuntimeOption()
+runtime_option.use_sophgo()
+
+model_file = args.model
+params_file = ""
+config_file = args.config_file
+
+model = fd.vision.segmentation.PaddleSegModel(
+    model_file,
+    params_file,
+    config_file,
+    runtime_option=runtime_option,
+    model_format=fd.ModelFormat.SOPHGO)
+
+# 预测图片分类结果
+im_org = cv2.imread(args.image)
+#bmodel 是静态模型，模型输入固定，这里设置为[512, 512]
+im = cv2.resize(im_org, [512, 512], interpolation=cv2.INTER_LINEAR)
+result = model.predict(im)
+print(result)
+
+# 预测结果可视化
+vis_im = fd.vision.vis_segmentation(im, result, weight=0.5)
+cv2.imwrite("sophgo_img.png", vis_im)