更换文档检测模型

2024-08-27 14:42:45 +08:00
parent aea6f19951
commit 1514e09c40
2072 changed files with 254336 additions and 4967 deletions
--- a/paddle_detection/deploy/fastdeploy/sophgo/README.md
+++ b/paddle_detection/deploy/fastdeploy/sophgo/README.md
@@ -0,0 +1,108 @@
+# PaddleDetection SOPHGO部署示例
+
+## 1. 支持模型列表
+
+目前SOPHGO支持如下模型的部署
+- [PP-YOLOE系列模型](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/ppyoloe)
+- [PicoDet系列模型](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/picodet)
+- [YOLOV8系列模型](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4)
+
+## 2. 准备PP-YOLOE YOLOV8或者PicoDet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:
+- Paddle动态图模型转换为ONNX模型，请参考[PaddleDetection导出模型](https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.4/deploy/EXPORT_MODEL.md).
+- ONNX模型转换bmodel模型的过程，请参考[TPU-MLIR](https://github.com/sophgo/tpu-mlir)
+
+## 3. 模型转换example
+
+PP-YOLOE YOLOV8和PicoDet模型转换过程类似，下面以ppyoloe_crn_s_300e_coco为例子,教大家如何转换Paddle模型到SOPHGO-TPU模型
+
+### 导出ONNX模型
+```shell
+#导出paddle模型
+python tools/export_model.py -c configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml --output_dir=output_inference -o weights=https://paddledet.bj.bcebos.com/models/ppyoloe_crn_s_300e_coco.pdparams
+
+#paddle模型转ONNX模型
+paddle2onnx --model_dir ppyoloe_crn_s_300e_coco \
+            --model_filename model.pdmodel \
+            --params_filename model.pdiparams \
+            --save_file ppyoloe_crn_s_300e_coco.onnx \
+            --enable_dev_version True
+
+#进入Paddle2ONNX文件夹，固定ONNX模型shape
+python -m paddle2onnx.optimize --input_model ppyoloe_crn_s_300e_coco.onnx \
+                                --output_model ppyoloe_crn_s_300e_coco.onnx \
+                                --input_shape_dict "{'image':[1,3,640,640]}"
+
+```
+### 导出bmodel模型
+
+以转化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)。
+## 4.	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
+```
+
+## 5.	ONNX模型转换为bmodel模型
+``` shell
+mkdir ppyoloe_crn_s_300e_coco && cd ppyoloe_crn_s_300e_coco
+
+# 下载测试图片，并将图片转换为npz格式
+wget https://gitee.com/paddlepaddle/PaddleDetection/raw/release/2.4/demo/000000014439.jpg
+
+#使用python获得模型转换所需要的npz文件
+im = cv2.imread(im)
+im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB)
+#[640 640]为ppyoloe_crn_s_300e_coco的输入大小
+im_scale_y = 640 / float(im.shape[0])
+im_scale_x = 640 / float(im.shape[1])
+inputs = {}
+inputs['image'] = np.array((im, )).astype('float32')
+inputs['scale_factor'] = np.array([im_scale_y, im_scale_x]).astype('float32')
+np.savez('inputs.npz', image = inputs['image'], scale_factor = inputs['scale_factor'])
+
+#放入onnx模型文件ppyoloe_crn_s_300e_coco.onnx
+
+mkdir workspace && cd workspace
+
+# 将ONNX模型转换为mlir模型
+model_transform.py \
+    --model_name ppyoloe_crn_s_300e_coco \
+    --model_def ../ppyoloe_crn_s_300e_coco.onnx \
+    --input_shapes [[1,3,640,640],[1,2]] \
+    --keep_aspect_ratio \
+    --pixel_format rgb \
+    --output_names p2o.Div.1,p2o.Concat.29 \
+    --test_input ../inputs.npz \
+    --test_result ppyoloe_crn_s_300e_coco_top_outputs.npz \
+    --mlir ppyoloe_crn_s_300e_coco.mlir
+```
+## 6. 注意
+**由于TPU-MLIR当前不支持后处理算法，所以需要查看后处理的输入作为网络的输出**  
+具体方法为：output_names需要通过[NETRO](https://netron.app/) 查看，网页中打开需要转换的ONNX模型，搜索NonMaxSuppression节点  
+查看INPUTS中boxes和scores的名字，这个两个名字就是我们所需的output_names  
+例如使用Netron可视化后，可以得到如下图片  
+![](https://user-images.githubusercontent.com/120167928/210939488-a37e6c8b-474c-4948-8362-2066ee7a2ecb.png)  
+找到蓝色方框标记的NonMaxSuppression节点，可以看到红色方框标记的两个节点名称为p2o.Div.1,p2o.Concat.29
+
+``` bash
+# 将mlir模型转换为BM1684x的F32 bmodel模型
+model_deploy.py \
+  --mlir ppyoloe_crn_s_300e_coco.mlir \
+  --quantize F32 \
+  --chip bm1684x \
+  --test_input ppyoloe_crn_s_300e_coco_in_f32.npz \
+  --test_reference ppyoloe_crn_s_300e_coco_top_outputs.npz \
+  --model ppyoloe_crn_s_300e_coco_1684x_f32.bmodel
+```
+最终获得可以在BM1684x上能够运行的bmodel模型ppyoloe_crn_s_300e_coco_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)。
+
+## 7. 详细的部署示例
+- [Cpp部署](./cpp)
+- [python部署](./python)
--- a/paddle_detection/deploy/fastdeploy/sophgo/cpp/CMakeLists.txt
+++ b/paddle_detection/deploy/fastdeploy/sophgo/cpp/CMakeLists.txt
@@ -0,0 +1,14 @@
+PROJECT(infer_demo C CXX)
+CMAKE_MINIMUM_REQUIRED (VERSION 3.10)
+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)
+
+include_directories(${FASTDEPLOY_INCS})
+include_directories(${FastDeploy_INCLUDE_DIRS})
+
+add_executable(infer_demo ${PROJECT_SOURCE_DIR}/infer.cc)
+target_link_libraries(infer_demo ${FASTDEPLOY_LIBS})
--- a/paddle_detection/deploy/fastdeploy/sophgo/cpp/README.md
+++ b/paddle_detection/deploy/fastdeploy/sophgo/cpp/README.md
@@ -0,0 +1,57 @@
+# PaddleDetection 算能 C++部署示例
+
+本目录下提供`infer.cc`,`快速完成 PP-YOLOE ,在SOPHGO BM1684x板子上加速部署的示例。PP-YOLOV8和 PicoDet的部署逻辑类似，只需要切换模型即可。
+
+## 1. 部署环境准备
+在部署前，需自行编译基于算能硬件的预测库，参考文档[算能硬件部署环境](https://github.com/PaddlePaddle/FastDeploy/blob/develop/docs/cn/build_and_install#算能硬件部署环境)
+
+## 2. 部署模型准备  
+在部署前，请准备好您所需要运行的推理模型，你可以选择使用[预导出的推理模型](../README.md)或者[自行导出PaddleDetection部署模型](../README.md)。
+
+## 3. 生成基本目录文件
+
+该例程由以下几个部分组成
+```text
+.
+├── CMakeLists.txt
+├── fastdeploy-sophgo  # 编译文件夹
+├── image  # 存放图片的文件夹
+├── infer.cc
+└── model  # 存放模型文件的文件夹
+```
+
+## 4. 运行部署示例
+
+### 4.1 编译并拷贝SDK到thirdpartys文件夹
+
+请参考[SOPHGO部署库编译](https://github.com/PaddlePaddle/FastDeploy/blob/develop/docs/cn/build_and_install/sophgo.md)仓库编译SDK，编译完成后，将在build目录下生成fastdeploy-sophgo目录.
+
+### 4.2 拷贝模型文件，以及配置文件至model文件夹
+将Paddle模型转换为SOPHGO bmodel模型，转换步骤参考[文档](../README.md)  
+将转换后的SOPHGO bmodel模型文件拷贝至model中
+
+### 4.3 准备测试图片至image文件夹
+```bash
+wget https://gitee.com/paddlepaddle/PaddleDetection/raw/release/2.4/demo/000000014439.jpg
+cp 000000014439.jpg ./images
+```
+
+### 4.4 编译example
+
+```bash
+cd build
+cmake .. -DFASTDEPLOY_INSTALL_DIR=${PWD}/fastdeploy-sophgo
+make
+```
+
+## 4.5 运行例程
+
+```bash
+#ppyoloe推理示例
+./infer_demo model images/000000014439.jpg
+```
+
+## 5. 更多指南
+- [FastDeploy部署PaddleDetection模型概览](../../)
+- [Python部署](../python)
+- [模型转换](../README.md)
--- a/paddle_detection/deploy/fastdeploy/sophgo/cpp/infer.cc
+++ b/paddle_detection/deploy/fastdeploy/sophgo/cpp/infer.cc
@@ -0,0 +1,60 @@
+// 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 <sys/time.h>
+
+#include <iostream>
+#include <string>
+
+#include "fastdeploy/vision.h"
+
+void SophgoInfer(const std::string& model_dir, const std::string& image_file) {
+  auto model_file = model_dir + "/ppyoloe_crn_s_300e_coco_1684x_f32.bmodel";
+  auto params_file = "";
+  auto config_file = model_dir + "/infer_cfg.yml";
+
+  auto option = fastdeploy::RuntimeOption();
+  option.UseSophgo();
+
+  auto format = fastdeploy::ModelFormat::SOPHGO;
+
+  auto model = fastdeploy::vision::detection::PPYOLOE(
+      model_file, params_file, config_file, option, format);
+
+  model.GetPostprocessor().ApplyNMS();
+
+  auto im = cv::imread(image_file);
+
+  fastdeploy::vision::DetectionResult res;
+  if (!model.Predict(&im, &res)) {
+    std::cerr << "Failed to predict." << std::endl;
+    return;
+  }
+
+  std::cout << res.Str() << std::endl;
+  auto vis_im = fastdeploy::vision::VisDetection(im, res, 0.5);
+  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, "
+           "e.g ./infer_demo ./model_dir ./test.jpeg"
+        << std::endl;
+    return -1;
+  }
+  SophgoInfer(argv[1], argv[2]);
+  return 0;
+}
--- a/paddle_detection/deploy/fastdeploy/sophgo/python/README.md
+++ b/paddle_detection/deploy/fastdeploy/sophgo/python/README.md
@@ -0,0 +1,30 @@
+# PaddleDetection Python部署示例
+
+## 1. 部署环境准备
+
+在部署前，需自行编译基于算能硬件的FastDeploy python wheel包并安装，参考文档[算能硬件部署环境](https://github.com/PaddlePaddle/FastDeploy/blob/develop/docs/cn/build_and_install#算能硬件部署环境)
+
+本目录下提供`infer.py`, 快速完成 PP-YOLOE ,在SOPHGO TPU上部署的示例，执行如下脚本即可完成。PP-YOLOV8和 PicoDet的部署逻辑类似，只需要切换模型即可。 
+
+## 2. 部署模型准备  
+在部署前，请准备好您所需要运行的推理模型，你可以选择使用[预导出的推理模型](../README.md)或者[自行导出PaddleDetection部署模型](../README.md)。
+
+```bash
+# 下载部署示例代码
+git clone https://github.com/PaddlePaddle/PaddleDetection.git
+cd PaddleDetection/deploy/fastdeploy/sophgo/python
+
+# 下载图片
+wget https://gitee.com/paddlepaddle/PaddleDetection/raw/release/2.4/demo/000000014439.jpg
+
+# 推理
+#ppyoloe推理示例
+python3 infer.py --model_file model/ppyoloe_crn_s_300e_coco_1684x_f32.bmodel --config_file model/infer_cfg.yml --image_file ./000000014439.jpg
+
+# 运行完成后返回结果如下所示
+可视化结果存储在sophgo_result.jpg中
+```
+
+## 3. 更多指南
+- [C++部署](../cpp)
+- [转换PP-YOLOE SOPHGO模型文档](../README.md)
--- a/paddle_detection/deploy/fastdeploy/sophgo/python/infer.py
+++ b/paddle_detection/deploy/fastdeploy/sophgo/python/infer.py
@@ -0,0 +1,59 @@
+# 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.
+import fastdeploy as fd
+import cv2
+import os
+
+
+def parse_arguments():
+    import argparse
+    import ast
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--model_file", required=True, help="Path of sophgo model.")
+    parser.add_argument("--config_file", required=True, help="Path of config.")
+    parser.add_argument(
+        "--image_file", type=str, required=True, help="Path of test image file.")
+    return parser.parse_args()
+
+
+if __name__ == "__main__":
+    args = parse_arguments()
+
+    model_file = args.model_file
+    params_file = ""
+    config_file = args.config_file
+
+    # setup runtime
+    runtime_option = fd.RuntimeOption()
+    runtime_option.use_sophgo()
+
+    model = fd.vision.detection.PPYOLOE(
+        model_file,
+        params_file,
+        config_file,
+        runtime_option=runtime_option,
+        model_format=fd.ModelFormat.SOPHGO)
+
+    model.postprocessor.apply_nms()
+
+    # predict
+    im = cv2.imread(args.image_file)
+    result = model.predict(im)
+    print(result)
+
+    # visualize
+    vis_im = fd.vision.vis_detection(im, result, score_threshold=0.5)
+    cv2.imwrite("sophgo_result.jpg", vis_im)
+    print("Visualized result save in ./sophgo_result.jpg")