[Quantization] Update quantized model deployment examples and update readme. (#377)

* Add PaddleOCR Support * Add PaddleOCR Support * Add PaddleOCRv3 Support * Add PaddleOCRv3 Support * Update README.md * Update README.md * Update README.md * Update README.md * Add PaddleOCRv3 Support * Add PaddleOCRv3 Supports * Add PaddleOCRv3 Suport * Fix Rec diff * Remove useless functions * Remove useless comments * Add PaddleOCRv2 Support * Add PaddleOCRv3 & PaddleOCRv2 Support * remove useless parameters * Add utils of sorting det boxes * Fix code naming convention * Fix code naming convention * Fix code naming convention * Fix bug in the Classify process * Imporve OCR Readme * Fix diff in Cls model * Update Model Download Link in Readme * Fix diff in PPOCRv2 * Improve OCR readme * Imporve OCR readme * Improve OCR readme * Improve OCR readme * Imporve OCR readme * Improve OCR readme * Fix conflict * Add readme for OCRResult * Improve OCR readme * Add OCRResult readme * Improve OCR readme * Improve OCR readme * Add Model Quantization Demo * Fix Model Quantization Readme * Fix Model Quantization Readme * Add the function to do PTQ quantization * Improve quant tools readme * Improve quant tool readme * Improve quant tool readme * Add PaddleInference-GPU for OCR Rec model * Add QAT method to fastdeploy-quantization tool * Remove examples/slim for now * Move configs folder * Add Quantization Support for Classification Model * Imporve ways of importing preprocess * Upload YOLO Benchmark on readme * Upload YOLO Benchmark on readme * Upload YOLO Benchmark on readme * Improve Quantization configs and readme * Add support for multi-inputs model * Add backends and params file for YOLOv7 * Add quantized model deployment support for YOLO series * Fix YOLOv5 quantize readme * Fix YOLO quantize readme * Fix YOLO quantize readme * Improve quantize YOLO readme * Improve quantize YOLO readme * Improve quantize YOLO readme * Improve quantize YOLO readme * Improve quantize YOLO readme * Fix bug, change Fronted to ModelFormat * Change Fronted to ModelFormat * Add examples to deploy quantized paddleclas models * Fix readme * Add quantize Readme * Add quantize Readme * Add quantize Readme * Modify readme of quantization tools * Modify readme of quantization tools * Improve quantization tools readme * Improve quantization readme * Improve PaddleClas quantized model deployment readme * Add PPYOLOE-l quantized deployment examples * Improve quantization tools readme * Improve Quantize Readme * Fix conflicts * Fix conflicts * improve readme * Improve quantization tools and readme * Improve quantization tools and readme * Add quantized deployment examples for PaddleSeg model * Fix cpp readme * Fix memory leak of reader_wrapper function * Fix model file name in PaddleClas quantization examples * Update Runtime and E2E benchmark * Update Runtime and E2E benchmark * Rename quantization tools to auto compression tools * Remove PPYOLOE data when deployed on MKLDNN * Fix readme * Support PPYOLOE with OR without NMS and update readme * Update Readme * Update configs and readme * Update configs and readme * Add Paddle-TensorRT backend in quantized model deploy examples * Support PPYOLOE+ series
2025-10-21 15:49:31 +08:00 · 2022-11-02 20:29:29 +08:00
parent 9437dec9f5
commit a231c9e7f3
53 changed files with 1514 additions and 521 deletions
--- a/examples/vision/segmentation/paddleseg/quantize/README.md
+++ b/examples/vision/segmentation/paddleseg/quantize/README.md
@@ -0,0 +1,36 @@
+# PaddleSeg 量化模型部署
+FastDeploy已支持部署量化模型,并提供一键模型自动化压缩的工具.
+用户可以使用一键模型自动化压缩工具,自行对模型量化后部署, 也可以直接下载FastDeploy提供的量化模型进行部署.
+
+## FastDeploy一键模型自动化压缩工具
+FastDeploy 提供了一键模型自动化压缩工具, 能够简单地通过输入一个配置文件, 对模型进行量化.
+详细教程请见: [一键模型自动化压缩工具](../../../../../tools/auto_compression/)
+注意: 推理量化后的分类模型仍然需要FP32模型文件夹下的deploy.yaml文件, 自行量化的模型文件夹内不包含此yaml文件, 用户从FP32模型文件夹下复制此yaml文件到量化后的模型文件夹内即可。
+
+## 下载量化完成的PaddleSeg模型
+用户也可以直接下载下表中的量化模型进行部署.(点击模型名字即可下载)
+
+Benchmark表格说明:
+- Rtuntime时延为模型在各种Runtime上的推理时延,包含CPU->GPU数据拷贝,GPU推理,GPU->CPU数据拷贝时间. 不包含模型各自的前后处理时间.
+- 端到端时延为模型在实际推理场景中的时延, 包含模型的前后处理.
+- 所测时延均为推理1000次后求得的平均值, 单位是毫秒.
+- INT8 + FP16 为在推理INT8量化模型的同时, 给Runtime 开启FP16推理选项
+- INT8 + FP16 + PM, 为在推理INT8量化模型和开启FP16的同时, 开启使用Pinned Memory的选项,可加速GPU->CPU数据拷贝的速度
+- 最大加速比, 为FP32时延除以INT8推理的最快时延,得到最大加速比.
+- 策略为量化蒸馏训练时, 采用少量无标签数据集训练得到量化模型, 并在全量验证集上验证精度, INT8精度并不代表最高的INT8精度.
+- CPU为Intel(R) Xeon(R) Gold 6271C, 所有测试中固定CPU线程数为1.  GPU为Tesla T4, TensorRT版本8.4.15.
+
+#### Runtime Benchmark
+| 模型                 |推理后端            |部署硬件    | FP32 Runtime时延   | INT8 Runtime时延 | INT8 + FP16 Runtime时延  | INT8+FP16+PM Runtime时延  | 最大加速比    | FP32 mIoU | INT8 mIoU | 量化方式   |
+| ------------------- | -----------------|-----------|  --------     |--------      |--------      | --------- |-------- |----- |----- |----- |
+| [PP-LiteSeg-T(STDC1)-cityscapes](https://bj.bcebos.com/paddlehub/fastdeploy/PP_LiteSeg_T_STDC1_cityscapes_without_argmax_infer_QAT_new.tar))  | Paddle Inference |    CPU    |     1138.04|   602.62 |None|None     |      1.89      |77.37 | 71.62 |量化蒸馏训练 |
+
+#### 端到端 Benchmark
+| 模型                 |推理后端            |部署硬件    | FP32 Runtime时延   | INT8 Runtime时延 | INT8 + FP16 Runtime时延  | INT8+FP16+PM Runtime时延  | 最大加速比    | FP32 mIoU | INT8 mIoU | 量化方式   |
+| ------------------- | -----------------|-----------|  --------     |--------      |--------      | --------- |-------- |----- |----- |----- |
+| [PP-LiteSeg-T(STDC1)-cityscapes](https://bj.bcebos.com/paddlehub/fastdeploy/PP_LiteSeg_T_STDC1_cityscapes_without_argmax_infer_QAT_new.tar))  | Paddle Inference |    CPU    |     4726.65|   4134.91|None|None     |      1.14      |77.37 | 71.62 |量化蒸馏训练 |
+
+## 详细部署文档
+
+- [Python部署](python)
+- [C++部署](cpp)
--- a/examples/vision/segmentation/paddleseg/quantize/cpp/CMakeLists.txt
+++ b/examples/vision/segmentation/paddleseg/quantize/cpp/CMakeLists.txt
@@ -0,0 +1,14 @@
+PROJECT(infer_demo C CXX)
+CMAKE_MINIMUM_REQUIRED (VERSION 3.12)
+
+# 指定下载解压后的fastdeploy库路径
+option(FASTDEPLOY_INSTALL_DIR "Path of downloaded fastdeploy sdk.")
+
+include(${FASTDEPLOY_INSTALL_DIR}/FastDeploy.cmake)
+
+# 添加FastDeploy依赖头文件
+include_directories(${FASTDEPLOY_INCS})
+
+add_executable(infer_demo ${PROJECT_SOURCE_DIR}/infer.cc)
+# 添加FastDeploy库依赖
+target_link_libraries(infer_demo ${FASTDEPLOY_LIBS})
--- a/examples/vision/segmentation/paddleseg/quantize/cpp/README.md
+++ b/examples/vision/segmentation/paddleseg/quantize/cpp/README.md
@@ -0,0 +1,30 @@
+# PaddleSeg 量化模型 C++部署示例
+本目录下提供的`infer.cc`,可以帮助用户快速完成PaddleSeg量化模型在CPU/GPU上的部署推理加速.
+
+## 部署准备
+### FastDeploy环境准备
+- 1. 软硬件环境满足要求，参考[FastDeploy环境要求](../../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md)  
+- 2. FastDeploy Python whl包安装，参考[FastDeploy Python安装](../../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md)
+
+### 量化模型准备
+- 1. 用户可以直接使用由FastDeploy提供的量化模型进行部署.
+- 2. 用户可以使用FastDeploy提供的[一键模型自动化压缩工具](../../tools/auto_compression/),自行进行模型量化, 并使用产出的量化模型进行部署.(注意: 推理量化后的分类模型仍然需要FP32模型文件夹下的deploy.yaml文件, 自行量化的模型文件夹内不包含此yaml文件, 用户从FP32模型文件夹下复制此yaml文件到量化后的模型文件夹内即可.)
+
+## 以量化后的PP_LiteSeg_T_STDC1_cityscapes模型为例, 进行部署
+在本目录执行如下命令即可完成编译,以及量化模型部署.
+```bash
+mkdir build
+cd build
+wget https://bj.bcebos.com/fastdeploy/release/cpp/fastdeploy-linux-x64-0.3.0.tgz
+tar xvf fastdeploy-linux-x64-0.3.0.tgz
+cmake .. -DFASTDEPLOY_INSTALL_DIR=${PWD}/fastdeploy-linux-x64-0.3.0
+make -j
+
+#下载FastDeloy提供的PP_LiteSeg_T_STDC1_cityscapes量化模型文件和测试图片
+wget https://bj.bcebos.com/paddlehub/fastdeploy/PP_LiteSeg_T_STDC1_cityscapes_without_argmax_infer_PTQ.tar
+tar -xvf PP_LiteSeg_T_STDC1_cityscapes_without_argmax_infer_PTQ.tar
+wget https://paddleseg.bj.bcebos.com/dygraph/demo/cityscapes_demo.png
+
+# 在CPU上使用Paddle-Inference推理量化模型
+./infer_demo PP_LiteSeg_T_STDC1_cityscapes_without_argmax_infer_PTQ cityscapes_demo.png 1
+```
--- a/examples/vision/segmentation/paddleseg/quantize/cpp/infer.cc
+++ b/examples/vision/segmentation/paddleseg/quantize/cpp/infer.cc
@@ -0,0 +1,100 @@
+// 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/vision.h"
+#ifdef WIN32
+const char sep = '\\';
+#else
+const char sep = '/';
+#endif
+
+void InitAndInfer(const std::string& model_dir, const std::string& image_file,
+                  const fastdeploy::RuntimeOption& option) {
+  auto model_file = model_dir + sep + "model.pdmodel";
+  auto params_file = model_dir + sep + "model.pdiparams";
+  auto config_file = model_dir + sep + "deploy.yaml";
+
+  auto model = fastdeploy::vision::segmentation::PaddleSegModel(
+      model_file, params_file, config_file,option);
+
+  assert(model.Initialized());
+
+  auto im = cv::imread(image_file);
+  auto im_bak = im.clone();
+
+  fastdeploy::vision::SegmentationResult res;
+  if (!model.Predict(&im, &res)) {
+    std::cerr << "Failed to predict." << std::endl;
+    return;
+  }
+
+  std::cout << res.Str() << 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 ./ppseg_model_dir ./test.jpeg 0"
+//         << std::endl;
+//     std::cout << "The data type of run_option is int, 0: run with cpu; 1: run "
+//                  "with gpu; 2: run with gpu and use tensorrt backend."
+//               << std::endl;
+//     return -1;
+//   }
+
+//   fastdeploy::RuntimeOption option;
+//   option.UseCpu();
+//   option.UsePaddleBackend();
+//   std::cout<<"Xyy-debug, enable Paddle Backend==!";
+
+//   std::string model_dir = argv[1];
+//   std::string test_image = argv[2];
+//   InitAndInfer(model_dir, test_image, option);
+//   return 0;
+// }
+
+int main(int argc, char* argv[]) {
+  if (argc < 4) {
+    std::cout << "Usage: infer_demo path/to/quant_model "
+                 "path/to/image "
+                 "run_option, "
+                 "e.g ./infer_demo ./ResNet50_vd_quant ./test.jpeg 0"
+              << std::endl;
+    std::cout << "The data type of run_option is int, 0: run on cpu with ORT "
+                 "backend; 1: run "
+                 "on gpu with TensorRT backend. "
+              << std::endl;
+    return -1;
+  }
+
+  fastdeploy::RuntimeOption option;
+  int flag = std::atoi(argv[3]);
+
+  if (flag == 0) {
+    option.UseCpu();
+    option.UseOrtBackend();
+    std::cout<<"Use ORT!"<<std::endl;
+  } else if (flag == 1) {
+    option.UseCpu();
+    option.UsePaddleBackend();
+    std::cout<<"Use PP!"<<std::endl;
+  }
+
+  std::string model_dir = argv[1];
+  std::string test_image = argv[2];
+  InitAndInfer(model_dir, test_image, option);
+  return 0;
+}
--- a/examples/vision/segmentation/paddleseg/quantize/python/README.md
+++ b/examples/vision/segmentation/paddleseg/quantize/python/README.md
@@ -0,0 +1,28 @@
+# PaddleSeg 量化模型 Python部署示例
+本目录下提供的`infer.py`,可以帮助用户快速完成PaddleSeg量化模型在CPU/GPU上的部署推理加速.
+
+## 部署准备
+### FastDeploy环境准备
+- 1. 软硬件环境满足要求，参考[FastDeploy环境要求](../../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md)  
+- 2. FastDeploy Python whl包安装，参考[FastDeploy Python安装](../../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md)
+
+### 量化模型准备
+- 1. 用户可以直接使用由FastDeploy提供的量化模型进行部署.
+- 2. 用户可以使用FastDeploy提供的[一键模型自动化压缩工具](../../tools/auto_compression/),自行进行模型量化, 并使用产出的量化模型进行部署.(注意: 推理量化后的分类模型仍然需要FP32模型文件夹下的deploy.yaml文件, 自行量化的模型文件夹内不包含此yaml文件, 用户从FP32模型文件夹下复制此yaml文件到量化后的模型文件夹内即可.)
+
+
+## 以量化后的PP_LiteSeg_T_STDC1_cityscapes模型为例, 进行部署
+```bash
+#下载部署示例代码
+git clone https://github.com/PaddlePaddle/FastDeploy.git
+cd examples/vision/segmentation/paddleseg/quantize/python
+
+#下载FastDeloy提供的PP_LiteSeg_T_STDC1_cityscapes量化模型文件和测试图片
+wget https://bj.bcebos.com/paddlehub/fastdeploy/PP_LiteSeg_T_STDC1_cityscapes_without_argmax_infer_PTQ.tar
+tar -xvf PP_LiteSeg_T_STDC1_cityscapes_without_argmax_infer_PTQ.tar
+wget https://paddleseg.bj.bcebos.com/dygraph/demo/cityscapes_demo.png
+
+# 在CPU上使用Paddle-Inference推理量化模型
+python infer.py --model PP_LiteSeg_T_STDC1_cityscapes_without_argmax_infer_QAT --image cityscapes_demo.png --device cpu --backend paddle
+
+```
--- a/examples/vision/segmentation/paddleseg/quantize/python/infer.py
+++ b/examples/vision/segmentation/paddleseg/quantize/python/infer.py
@@ -0,0 +1,76 @@
+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 PaddleSeg model.")
+    parser.add_argument(
+        "--image", required=True, help="Path of test image file.")
+    parser.add_argument(
+        "--device",
+        type=str,
+        default='cpu',
+        help="Type of inference device, support 'cpu' or 'gpu'.")
+    parser.add_argument(
+        "--backend",
+        type=str,
+        default="default",
+        help="Type of inference backend, support ort/trt/paddle/openvino, default 'openvino' for cpu, 'tensorrt' for gpu"
+    )
+    parser.add_argument(
+        "--device_id",
+        type=int,
+        default=0,
+        help="Define which GPU card used to run model.")
+    parser.add_argument(
+        "--cpu_thread_num",
+        type=int,
+        default=9,
+        help="Number of threads while inference on CPU.")
+    return parser.parse_args()
+
+
+def build_option(args):
+    option = fd.RuntimeOption()
+    if args.device.lower() == "gpu":
+        option.use_gpu(0)
+
+    option.set_cpu_thread_num(args.cpu_thread_num)
+
+    if args.backend.lower() == "trt":
+        assert args.device.lower(
+        ) == "gpu", "TensorRT backend require inferences on device GPU."
+        option.use_trt_backend()
+        option.set_trt_cache_file(os.path.join(args.model, "model.trt"))
+        option.set_trt_input_shape("x", [1, 3, 256, 256], [1, 3, 1024, 1024],
+                                   [1, 3, 2048, 2048])
+    elif args.backend.lower() == "ort":
+        option.use_ort_backend()
+    elif args.backend.lower() == "paddle":
+        option.use_paddle_backend()
+    elif args.backend.lower() == "openvino":
+        assert args.device.lower(
+        ) == "cpu", "OpenVINO backend require inference on device CPU."
+        option.use_openvino_backend()
+    return option
+
+
+args = parse_arguments()
+
+# 配置runtime，加载模型
+runtime_option = build_option(args)
+model_file = os.path.join(args.model, "model.pdmodel")
+params_file = os.path.join(args.model, "model.pdiparams")
+config_file = os.path.join(args.model, "deploy.yaml")
+model = fd.vision.segmentation.PaddleSegModel(
+    model_file, params_file, config_file, runtime_option=runtime_option)
+
+# 预测图片检测结果
+im = cv2.imread(args.image)
+result = model.predict(im.copy())
+print(result)