apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 08:37:06 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

modify2

Browse Source
This commit is contained in:
Winter
2022-11-08 10:52:47 +08:00
committed by WinterGeng
parent 09227b2ffc bdf40e9da5
commit 277c86e612
47 changed files with 195 additions and 110 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
README.md
View File

@@ -1 +1 @@
README_EN.md
README_CN.md

10
README_CN.md
View File

@@ -1,4 +1,4 @@
[English](README.md) | 简体中文
[English](README_EN.md) | 简体中文
![⚡️FastDeploy](https://user-images.githubusercontent.com/31974251/185771818-5d4423cd-c94c-4a49-9894-bc7a8d1c29d0.png)
@@ -18,13 +18,13 @@
**⚡️FastDeploy**是一款**易用高效**的推理部署开发套件。覆盖业界🔥**热门CV、NLP、Speech的AI模型**并提供📦**开箱即用**的部署体验，包括图像分类、目标检测、图像分割、人脸检测、人脸识别、人体关键点识别、文字识别、语义理解等多任务，满足开发者**多场景**，**多硬件**、**多平台**的产业部署需求。
| [Object Detection](examples/vision) | [3D Object Detection](https://github.com/PaddlePaddle/FastDeploy/issues/6) | [Semantic Segmentation](examples/vision/segmentation/paddleseg) | [Potrait Segmentation](examples/vision/segmentation/paddleseg) |
| [Object Detection](examples/vision/detection) | [3D Object Detection](https://github.com/PaddlePaddle/FastDeploy/issues/6) | [Semantic Segmentation](examples/vision/segmentation/paddleseg) | [Potrait Segmentation](examples/vision/segmentation/paddleseg) |
|:---------------------------------------------------------------------------------------------------------:|:---------------------------------------------------------------------------------------------------------:|:--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:|:-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:|
| <img src='https://user-images.githubusercontent.com/54695910/188054680-2f8d1952-c120-4b67-88fc-7d2d7d2378b4.gif' height="126px" width="190px"> | <img src='https://user-images.githubusercontent.com/54695910/188270227-1a4671b3-0123-46ab-8d0f-0e4132ae8ec0.gif' height="126px" width="190px"> | <img src='https://user-images.githubusercontent.com/54695910/188054711-6119f0e7-d741-43b1-b273-9493d103d49f.gif' height="126px" width="190px"> | <img src='https://user-images.githubusercontent.com/54695910/188054718-6395321c-8937-4fa0-881c-5b20deb92aaa.gif' height="126px" width="190px"> |
| [**Image Matting**](examples/vision/matting) | [**Real-Time Matting**](examples/vision/matting) | [**OCR**](examples/vision/ocr) |[**Face Alignment**](examples/vision/ocr)
| [**Image Matting**](examples/vision/matting) | [**Real-Time Matting**](examples/vision/matting) | [**OCR**](examples/vision/ocr) |[**Face Alignment**](examples/vision/facealign)
| <img src='https://user-images.githubusercontent.com/54695910/188058231-a5fe1ce1-0a38-460f-9582-e0b881514908.gif' height="126px" width="190px"> |<img src='https://user-images.githubusercontent.com/54695910/188054691-e4cb1a70-09fe-4691-bc62-5552d50bd853.gif' height="126px" width="190px"> | <img src='https://user-images.githubusercontent.com/54695910/188054669-a85996ba-f7f3-4646-ae1f-3b7e3e353e7d.gif' height="126px" width="190px" > |<img src='https://user-images.githubusercontent.com/54695910/188059460-9845e717-c30a-4252-bd80-b7f6d4cf30cb.png' height="126px" width="190px"> |
| [**Pose Estimation**](examples/vision/keypointdetection) | [**Behavior Recognition**](https://github.com/PaddlePaddle/FastDeploy/issues/6) | [**NLP**](examples/text) |[**Speech**](examples/audio/pp-tts)
| <img src='https://user-images.githubusercontent.com/54695910/188054671-394db8dd-537c-42b1-9d90-468d7ad1530e.gif' height="126px" width="190px"> | <img src='https://user-images.githubusercontent.com/48054808/173034825-623e4f78-22a5-4f14-9b83-dc47aa868478.gif' height="126px" width="190px"> | <img src='https://user-images.githubusercontent.com/54695910/200162475-f5d85d70-18fb-4930-8e7e-9ca065c1d618.gif' height="126px" width="190px"> | <p align="left">**input** ：早上好，今天是2020<br>/10/29，最低温度是-3°C。<br><br> <p align="left">**output**: [<img src="https://user-images.githubusercontent.com/54695910/200161645-871e08da-5a31-4736-879c-a88bb171a676.png" width="170" style="max-width: 100%;">](https://paddlespeech.bj.bcebos.com/Parakeet/docs/demos/parakeet_espnet_fs2_pwg_demo/tn_g2p/parakeet/001.wav)</p>|
| <img src='https://user-images.githubusercontent.com/54695910/188054671-394db8dd-537c-42b1-9d90-468d7ad1530e.gif' height="126px" width="190px"> | <img src='https://user-images.githubusercontent.com/48054808/173034825-623e4f78-22a5-4f14-9b83-dc47aa868478.gif' height="126px" width="190px"> | <img src='https://user-images.githubusercontent.com/54695910/200162475-f5d85d70-18fb-4930-8e7e-9ca065c1d618.gif' height="126px" width="190px"> | <p align="left">**input** ：早上好今天是2020<br>/10/29，最低温度是-3°C。<br><br> <p align="left">**output**: [<img src="https://user-images.githubusercontent.com/54695910/200161645-871e08da-5a31-4736-879c-a88bb171a676.png" width="170" style="max-width: 100%;">](https://paddlespeech.bj.bcebos.com/Parakeet/docs/demos/parakeet_espnet_fs2_pwg_demo/tn_g2p/parakeet/001.wav)</p>|
## 近期更新
@@ -40,7 +40,7 @@
- **🖥️ 服务端部署：支持推理速度更快的后端，支持更多的模型**
- 集成 Paddle Inference TensorRT后端，并保证其使用与Paddle Inference、TensorRT、OpenVINO、ONNX Runtime、Paddle Lite等一致的开发体验；
- 支持并测试 Graphcore IPU 通过 Paddle Inference后端;
- 优化[一键模型量化工具](tools/quantization)，支持YOLOv7、YOLOv6、YOLOv5等视觉模型，在CPU和GPU推理速度可提升1.5～2倍；
- 优化[一键模型自动化压缩工具](./tools/auto_compression)，支持YOLOv7、YOLOv6、YOLOv5等视觉模型，在CPU和GPU推理速度可提升1.5～2倍；
- 新增 [PP-Tracking](./examples/vision/tracking/pptracking) 和 [RobustVideoMatting](./examples/vision/matting) 等模型；
- 🔥 **2022.10.24：Release FastDeploy [release v0.4.0](https://github.com/PaddlePaddle/FastDeploy/tree/release/0.4.0)**

6
README_EN.md
View File

@@ -20,13 +20,13 @@ English | [简体中文](README_CN.md)
**⚡️FastDeploy** is an **accessible and efficient** deployment Development Toolkit. It covers 🔥**critical CV、NLP、Speech AI models** in the industry and provides 📦**out-of-the-box** deployment experience. It covers image classification, object detection, image segmentation, face detection, face recognition, human keypoint detection, OCR, semantic understanding and other tasks to meet developers' industrial deployment needs for **multi-scenario**, **multi-hardware** and **multi-platform** .
| [Object Detection](examples/vision) | [3D Object Detection](https://github.com/PaddlePaddle/FastDeploy/issues/6) | [Semantic Segmentation](examples/vision/segmentation/paddleseg) | [Potrait Segmentation](examples/vision/segmentation/paddleseg) |
| [Object Detection](examples/vision/detection) | [3D Object Detection](https://github.com/PaddlePaddle/FastDeploy/issues/6) | [Semantic Segmentation](examples/vision/segmentation/paddleseg) | [Potrait Segmentation](examples/vision/segmentation/paddleseg) |
|:---------------------------------------------------------------------------------------------------------:|:---------------------------------------------------------------------------------------------------------:|:--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:|:-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:|
| <img src='https://user-images.githubusercontent.com/54695910/188054680-2f8d1952-c120-4b67-88fc-7d2d7d2378b4.gif' height="126px" width="190px"> | <img src='https://user-images.githubusercontent.com/54695910/188270227-1a4671b3-0123-46ab-8d0f-0e4132ae8ec0.gif' height="126px" width="190px"> | <img src='https://user-images.githubusercontent.com/54695910/188054711-6119f0e7-d741-43b1-b273-9493d103d49f.gif' height="126px" width="190px"> | <img src='https://user-images.githubusercontent.com/54695910/188054718-6395321c-8937-4fa0-881c-5b20deb92aaa.gif' height="126px" width="190px"> |
| [**Image Matting**](examples/vision/matting) | [**Real-Time Matting**](examples/vision/matting) | [**OCR**](examples/vision/ocr) |[**Face Alignment**](examples/vision/ocr)
| [**Image Matting**](examples/vision/matting) | [**Real-Time Matting**](examples/vision/matting) | [**OCR**](examples/vision/ocr) |[**Face Alignment**](examples/vision/facealign)
| <img src='https://user-images.githubusercontent.com/54695910/188058231-a5fe1ce1-0a38-460f-9582-e0b881514908.gif' height="126px" width="190px"> |<img src='https://user-images.githubusercontent.com/54695910/188054691-e4cb1a70-09fe-4691-bc62-5552d50bd853.gif' height="126px" width="190px"> | <img src='https://user-images.githubusercontent.com/54695910/188054669-a85996ba-f7f3-4646-ae1f-3b7e3e353e7d.gif' height="126px" width="190px" > |<img src='https://user-images.githubusercontent.com/54695910/188059460-9845e717-c30a-4252-bd80-b7f6d4cf30cb.png' height="126px" width="190px"> |
| [**Pose Estimation**](examples/vision/keypointdetection) | [**Behavior Recognition**](https://github.com/PaddlePaddle/FastDeploy/issues/6) | [**NLP**](examples/text) |[**Speech**](examples/audio/pp-tts)
| <img src='https://user-images.githubusercontent.com/54695910/188054671-394db8dd-537c-42b1-9d90-468d7ad1530e.gif' height="126px" width="190px"> | <img src='https://user-images.githubusercontent.com/48054808/173034825-623e4f78-22a5-4f14-9b83-dc47aa868478.gif' height="126px" width="190px"> | <img src='https://user-images.githubusercontent.com/54695910/200162475-f5d85d70-18fb-4930-8e7e-9ca065c1d618.gif' height="126px" width="190px"> | <p align="left">**input** ：Life was like a box of chocolates, you never know what you're gonna get.<br> <p align="left">**output**: [<img src="https://user-images.githubusercontent.com/54695910/200161645-871e08da-5a31-4736-879c-a88bb171a676.png" width="170" style="max-width: 100%;">](https://paddlespeech.bj.bcebos.com/Parakeet/docs/demos/tacotron2_ljspeech_waveflow_samples_0.2/sentence_1.wav)</p>|
| <img src='https://user-images.githubusercontent.com/54695910/188054671-394db8dd-537c-42b1-9d90-468d7ad1530e.gif' height="126px" width="190px"> | <img src='https://user-images.githubusercontent.com/48054808/173034825-623e4f78-22a5-4f14-9b83-dc47aa868478.gif' height="126px" width="190px"> | <img src='https://user-images.githubusercontent.com/54695910/200162475-f5d85d70-18fb-4930-8e7e-9ca065c1d618.gif' height="126px" width="190px"> | <p align="left">**input** ：早上好今天是2020<br>/10/29，最低温度是-3°C。<br><br> <p align="left">**output**: [<img src="https://user-images.githubusercontent.com/54695910/200161645-871e08da-5a31-4736-879c-a88bb171a676.png" width="170" style="max-width: 100%;">](https://paddlespeech.bj.bcebos.com/Parakeet/docs/demos/parakeet_espnet_fs2_pwg_demo/tn_g2p/parakeet/001.wav)</p>|
## 📣 Recent Updates

2
docs/api_docs/python/README.md
View File

@@ -2,7 +2,7 @@
This directory help to generate Python API documents for FastDeploy.
1. First, to generate the latest api documents, you need to install the latest FastDeploy, refer [build and install](en/build_and_install) to build FastDeploy python wheel package with the latest code.
1. First, to generate the latest api documents, you need to install the latest FastDeploy, refer [build and install](../../cn/build_and_install) to build FastDeploy python wheel package with the latest code.
2. After installed FastDeploy in your python environment, there are some dependencies need to install, execute command `pip install -r requirements.txt` in this directory
3. Execute command `make html` to generate API documents

2
docs/cn/build_and_install/android.md
View File

@@ -102,4 +102,4 @@ make install
如何使用FastDeploy Android C++ SDK 请参考使用案例文档：
- [图像分类Android使用文档](../../../examples/vision/classification/paddleclas/android/README.md)
- [目标检测Android使用文档](../../../examples/vision/detection/paddledetection/android/README.md)
- [在 Android 通过 JNI 中使用 FastDeploy C++ SDK](../../../../../docs/cn/faq/use_cpp_sdk_on_android.md)
- [在 Android 通过 JNI 中使用 FastDeploy C++ SDK](../../cn/faq/use_cpp_sdk_on_android.md)

2
docs/cn/faq/use_sdk_on_windows.md
View File

@@ -218,7 +218,7 @@ D:\qiuyanjun\fastdeploy_test\infer_ppyoloe\x64\Release\infer_ppyoloe.exe
![image](https://user-images.githubusercontent.com/31974251/192144782-79bccf8f-65d0-4f22-9f41-81751c530319.png)
（2）其中infer_ppyoloe.cpp的代码可以直接从examples中的代码拷贝过来：
- [examples/vision/detection/paddledetection/cpp/infer_ppyoloe.cc](../../examples/vision/detection/paddledetection/cpp/infer_ppyoloe.cc)
- [examples/vision/detection/paddledetection/cpp/infer_ppyoloe.cc](../../../examples/vision/detection/paddledetection/cpp/infer_ppyoloe.cc)
（3）CMakeLists.txt主要包括配置FastDeploy C++ SDK的路径，如果是GPU版本的SDK，还需要配置CUDA_DIRECTORY为CUDA的安装路径，CMakeLists.txt的配置如下：

8
docs/en/faq/use_sdk_on_windows.md
View File

@@ -179,7 +179,7 @@ D:\qiuyanjun\fastdeploy_build\built\fastdeploy-win-x64-gpu-0.2.1\third_libs\inst
![image](https://user-images.githubusercontent.com/31974251/192827842-1f05d435-8a3e-492b-a3b7-d5e88f85f814.png)
Compile successfully, you can see the exe saved in：
Compile successfully, you can see the exe saved in：
```bat
D:\qiuyanjun\fastdeploy_test\infer_ppyoloe\x64\Release\infer_ppyoloe.exe
@@ -221,7 +221,7 @@ This section is for CMake users and describes how to create CMake projects in Vi
![image](https://user-images.githubusercontent.com/31974251/192144782-79bccf8f-65d0-4f22-9f41-81751c530319.png)
（2）The code of infer_ppyoloe.cpp can be copied directly from the code in examples：
- [examples/vision/detection/paddledetection/cpp/infer_ppyoloe.cc](../../examples/vision/detection/paddledetection/cpp/infer_ppyoloe.cc)
- [examples/vision/detection/paddledetection/cpp/infer_ppyoloe.cc](../../../examples/vision/detection/paddledetection/cpp/infer_ppyoloe.cc)
（3）CMakeLists.txt mainly includes the configuration of the path of FastDeploy C++ SDK, if it is the GPU version of the SDK, you also need to configure CUDA_DIRECTORY as the installation path of CUDA, the configuration of CMakeLists.txt is as follows：
@@ -361,7 +361,7 @@ A brief description of the usage is as follows.
#### 4.1.2 fastdeploy_init.bat View all dll, lib and include paths in the SDK
<div id="CommandLineDeps12"></div>
Go to the root directory of the SDK and run the show command to view all the dll, lib and include paths in the SDK. In the following command, %cd% means the current directory (the root directory of the SDK).
Go to the root directory of the SDK and run the show command to view all the dll, lib and include paths in the SDK. In the following command, %cd% means the current directory (the root directory of the SDK).
```bat
D:\path-to-fastdeploy-sdk-dir>fastdeploy_init.bat show %cd%
@@ -504,7 +504,7 @@ copy /Y %FASTDEPLOY_HOME%\third_libs\install\yaml-cpp\lib\*.dll Release\
copy /Y %FASTDEPLOY_HOME%\third_libs\install\openvino\bin\*.dll Release\
copy /Y %FASTDEPLOY_HOME%\third_libs\install\openvino\bin\*.xml Release\
copy /Y %FASTDEPLOY_HOME%\third_libs\install\openvino\3rdparty\tbb\bin\*.dll Release\
```
```
Note that if you compile the latest SDK or version >0.2.1 by yourself, the opencv and openvino directory structure has changed and the path needs to be modified appropriately. For example：
```bat
copy /Y %FASTDEPLOY_HOME%\third_libs\install\opencv\build\x64\vc15\bin\*.dll Release\

2
docs/en/quantize.md
View File

@@ -27,7 +27,7 @@ FastDeploy基于PaddleSlim, 集成了一键模型量化的工具, 同时, FastDe
### 用户使用FastDeploy一键模型量化工具来量化模型
Fastdeploy基于PaddleSlim, 为用户提供了一键模型量化的工具，请参考如下文档进行模型量化.
- [FastDeploy 一键模型量化](../../tools/quantization/)
- [FastDeploy 一键模型量化](../../tools/auto_compression/)
当用户获得产出的量化模型之后，即可以使用FastDeploy来部署量化模型.

28
examples/runtime/README.md
View File

@@ -1,16 +1,16 @@
# FastDeploy Runtime推理示例
# FastDeploy Runtime examples
| 示例代码 | 编程语言 | 说明 |
| Example Code | Program Language | Description |
| :------- | :------- | :---- |
| python/infer_paddle_paddle_inference.py | Python | paddle模型通过paddle inference在cpu/gpu上的推理 |
| python/infer_paddle_tensorrt.py | Python | paddle模型通过tensorrt在gpu上的推理 |
| python/infer_paddle_openvino.py | Python | paddle模型通过openvino在cpu上的推理 |
| python/infer_paddle_onnxruntime.py | Python | paddle模型通过onnx runtime在cpu/gpu上的推理 |
| python/infer_onnx_openvino.py | Python | onnx模型通过openvino在cpu上的推理 |
| python/infer_onnx_tensorrt.py | Python | onnx模型通过tensorrt在gpu上的推理 |
| cpp/infer_paddle_paddle_inference.cc | C++ | paddle模型通过paddle inference在cpu/gpu上的推理 |
| cpp/infer_paddle_tensorrt.cc | C++ | paddle模型通过tensorrt在gpu上的推理 |
| cpp/infer_paddle_openvino.cc | C++ | paddle模型通过openvino在cpu上的推理 |
| cpp/infer_paddle_onnxruntime.cc | C++ | paddle模型通过onnx runtime在cpu/gpu上的推理 |
| cpp/infer_onnx_openvino.cc | C++ | onnx模型通过openvino在cpu上的推理 |
| cpp/infer_onnx_tensorrt.cc | C++ | onnx模型通过tensorrt在gpu上的推理 |
| python/infer_paddle_paddle_inference.py | Python | Deploy Paddle model with Paddle Inference(CPU/GPU) |
| python/infer_paddle_tensorrt.py | Python | Deploy Paddle model with TensorRT(GPU) |
| python/infer_paddle_openvino.py | Python | Deploy Paddle model with OpenVINO(CPU) |
| python/infer_paddle_onnxruntime.py | Python | Deploy Paddle model with ONNX Runtime(CPU/GPU) |
| python/infer_onnx_openvino.py | Python | Deploy ONNX model with OpenVINO(CPU) |
| python/infer_onnx_tensorrt.py | Python | Deploy ONNX model with TensorRT(GPU) |
| cpp/infer_paddle_paddle_inference.cc | C++ | Deploy Paddle model with Paddle Inference(CPU/GPU) |
| cpp/infer_paddle_tensorrt.cc | C++ | Deploy Paddle model with TensorRT(GPU) |
| cpp/infer_paddle_openvino.cc | C++ | Deploy Paddle model with OpenVINO(CPU |
| cpp/infer_paddle_onnxruntime.cc | C++ | Deploy Paddle model with ONNX Runtime(CPU/GPU) |
| cpp/infer_onnx_openvino.cc | C++ | Deploy ONNX model with OpenVINO(CPU) |
| cpp/infer_onnx_tensorrt.cc | C++ | Deploy ONNX model with TensorRT(GPU) |

2
examples/text/ernie-3.0/serving/README.md
View File

@@ -168,4 +168,4 @@ entity: 华夏 label: LOC pos: [14, 15]
## 配置修改
当前分类任务(ernie_seqcls_model/config.pbtxt)默认配置在CPU上运行OpenVINO引擎; 序列标注任务默认配置在GPU上运行Paddle引擎。如果要在CPU/GPU或其他推理引擎上运行, 需要修改配置，详情请参考[配置文档](../../../../../serving/docs/zh_CN/model_configuration.md)
当前分类任务(ernie_seqcls_model/config.pbtxt)默认配置在CPU上运行OpenVINO引擎; 序列标注任务默认配置在GPU上运行Paddle引擎。如果要在CPU/GPU或其他推理引擎上运行, 需要修改配置，详情请参考[配置文档](../../../../serving/docs/zh_CN/model_configuration.md)

2
examples/vision/README.md
View File

@@ -30,4 +30,4 @@ FastDeploy针对飞桨的视觉套件，以及外部热门模型，提供端到
- 加载模型
- 调用`predict`接口
FastDeploy在各视觉模型部署时，也支持一键切换后端推理引擎，详情参阅[如何切换模型推理引擎](../../docs/runtime/how_to_change_backend.md)。
FastDeploy在各视觉模型部署时，也支持一键切换后端推理引擎，详情参阅[如何切换模型推理引擎](../../docs/cn/faq/how_to_change_backend.md)。

2
examples/vision/classification/paddleclas/quantize/cpp/README.md
View File

@@ -8,7 +8,7 @@
### 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inference_cls.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.)
- 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inference_cls.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ResNet50_Vd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.

2
examples/vision/classification/paddleclas/quantize/python/README.md
View File

@@ -8,7 +8,7 @@
### 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inference_cls.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.)
- 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inference_cls.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ResNet50_Vd模型为例, 进行部署

3
examples/vision/classification/paddleclas/web/README.md
View File

@@ -6,7 +6,7 @@
## 前端部署图像分类模型
图像分类模型web demo使用[**参考文档**](../../../../examples/application/js/web_demo)
图像分类模型web demo使用[**参考文档**](../../../../application/js/web_demo/)
## MobileNet js接口
@@ -34,4 +34,3 @@ console.log(res);
- [PaddleClas模型 python部署](../../paddleclas/python/)
- [PaddleClas模型 C++部署](../cpp/)

8
examples/vision/classification/resnet/cpp/README.md
View File

@@ -4,8 +4,8 @@
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/environment.md)
- 2. 根据开发环境，下载预编译部署库和samples代码，参考[FastDeploy预编译库](../../../../../docs/quick_start)
- 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. 根据开发环境，下载预编译部署库和samples代码，参考[FastDeploy预编译库](../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md)
以Linux上 ResNet50 推理为例，在本目录执行如下命令即可完成编译测试
@@ -33,7 +33,7 @@ wget https://gitee.com/paddlepaddle/PaddleClas/raw/release/2.4/deploy/images/Ima
```
以上命令只适用于Linux或MacOS, Windows下SDK的使用方式请参考:
- [如何在Windows中使用FastDeploy C++ SDK](../../../../../docs/compile/how_to_use_sdk_on_windows.md)
- [如何在Windows中使用FastDeploy C++ SDK](../../../../../docs/cn/faq/use_sdk_on_windows.md)
## ResNet C++接口
@@ -74,4 +74,4 @@ fastdeploy::vision::classification::ResNet(
- [模型介绍](../../)
- [Python部署](../python)
- [视觉模型预测结果](../../../../../docs/api/vision_results/)
- [如何切换模型推理后端引擎](../../../../../docs/runtime/how_to_change_backend.md)
- [如何切换模型推理后端引擎](../../../../../docs/cn/faq/how_to_change_backend.md)

6
examples/vision/classification/resnet/python/README.md
View File

@@ -2,8 +2,8 @@
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/environment.md)
- 2. FastDeploy Python whl包安装，参考[FastDeploy Python安装](../../../../../docs/quick_start)
- 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)
本目录下提供`infer.py`快速完成ResNet50_vd在CPU/GPU，以及GPU上通过TensorRT加速部署的示例。执行如下脚本即可完成
@@ -69,4 +69,4 @@ fd.vision.classification.ResNet(model_file, params_file, runtime_option=None, mo
- [ResNet 模型介绍](..)
- [ResNet C++部署](../cpp)
- [模型预测结果说明](../../../../../docs/api/vision_results/)
- [如何切换模型推理后端引擎](../../../../../docs/runtime/how_to_change_backend.md)
- [如何切换模型推理后端引擎](../../../../../docs/cn/faq/how_to_change_backend.md)

2
examples/vision/detection/paddledetection/quantize/cpp/README.md
View File

@@ -9,7 +9,7 @@
### 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infer_cfg.yml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.)
- 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infer_cfg.yml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-YOLOE-l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.

2
examples/vision/detection/paddledetection/quantize/python/README.md
View File

@@ -8,7 +8,7 @@
### 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infer_cfg.yml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.)
- 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infer_cfg.yml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-YOLOE-l模型为例, 进行部署

2
examples/vision/detection/yolov5/quantize/cpp/README.md
View File

@@ -9,7 +9,7 @@
### 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.
- 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.
## 以量化后的YOLOv5s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.

2
examples/vision/detection/yolov5/quantize/python/README.md
View File

@@ -8,7 +8,7 @@
### 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.
- 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.
## 以量化后的YOLOv5s模型为例, 进行部署

2
examples/vision/detection/yolov6/quantize/cpp/README.md
View File

@@ -9,7 +9,7 @@
### 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.
- 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.
## 以量化后的YOLOv6s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.

2
examples/vision/detection/yolov6/quantize/python/README.md
View File

@@ -8,7 +8,7 @@
### 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.
- 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.
## 以量化后的YOLOv6s模型为例, 进行部署
```bash

20
examples/vision/detection/yolov7/python/README_EN.md
View File

@@ -4,8 +4,8 @@ English | [简体中文](README.md)
Two steps before deployment:
- 1. The hardware and software environment meets the requirements. Please refer to [FastDeploy Environment Requirements](../../../../../docs/docs_en/environment.md)
- 2. Install FastDeploy Python whl package. Please refer to [FastDeploy Python Installation](../../../../../docs/docs_en/quick_start)
- 1. The hardware and software environment meets the requirements. Please refer to [FastDeploy Environment Requirements](../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md)
- 2. Install FastDeploy Python whl package. Please refer to [FastDeploy Python Installation](../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md)
This doc provides a quick `infer.py` demo of YOLOv7 deployment on CPU/GPU, and accelerated GPU deployment by TensorRT. Run the following command:
@@ -21,7 +21,7 @@ wget https://gitee.com/paddlepaddle/PaddleDetection/raw/release/2.4/demo/0000000
# CPU Inference
python infer.py --model yolov7.onnx --image 000000014439.jpg --device cpu
# GPU
# GPU
python infer.py --model yolov7.onnx --image 000000014439.jpg --device gpu
# GPU上使用TensorRT推理
python infer.py --model yolov7.onnx --image 000000014439.jpg --device gpu --use_trt True
@@ -51,18 +51,18 @@ YOLOv7 model loading and initialisation, with model_file being the exported ONNX
> ```python
> YOLOv7.predict(image_data, conf_threshold=0.25, nms_iou_threshold=0.5)
> ```
>
>
> Model prediction interface with direct output of detection results from the image input.
>
>
> **Parameters**
>
>
> > * **image_data**(np.ndarray): Input image. Images need to be in HWC or BGR format
> > * **conf_threshold**(float): Filter threshold for detection box confidence
> > * **nms_iou_threshold**(float): iou thresholds during NMS processing
> **Return**
>
> > Return to`fastdeploy.vision.DetectionResult`Struct. For more details, please refer to [Vision Model Results](../../../../../docs/docs_en/api/vision_results/)
>
> > Return to`fastdeploy.vision.DetectionResult`Struct. For more details, please refer to [Vision Model Results](../../../../../docs/api/vision_results/)
### Class Member Variables
@@ -80,5 +80,5 @@ Users can modify the following pre-processing parameters for their needs. This w
- [YOLOv7 Model Introduction](..)
- [YOLOv7 C++ Deployment](../cpp)
- [Vision Model Results](../../../../../docs/docs_en/api/vision_results/)
- [how to change inference backend](../../../../../docs/docs_en/runtime/how_to_change_inference_backend.md)
- [Vision Model Results](../../../../../docs/api/vision_results/)
- [how to change inference backend](../../../../../docs/en/faq/how_to_change_backend.md)

2
examples/vision/detection/yolov7/quantize/cpp/README.md
View File

@@ -9,7 +9,7 @@
### 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.
- 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.
## 以量化后的YOLOv7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.

2
examples/vision/detection/yolov7/quantize/python/README.md
View File

@@ -8,7 +8,7 @@
### 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.
- 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.
## 以量化后的YOLOv7模型为例, 进行部署
```bash

2
examples/vision/keypointdetection/det_keypoint_unite/python/README.md
View File

@@ -71,4 +71,4 @@ PPTinyPosePipeline模型加载和初始化，其中det_model是使用`fd.vision.
- [Pipeline 模型介绍](..)
- [Pipeline C++部署](../cpp)
- [模型预测结果说明](../../../../../docs/api/vision_results/)
- [如何切换模型推理后端引擎](../../../../../docs/runtime/how_to_change_backend.md)
- [如何切换模型推理后端引擎](../../../../../docs/cn/faq/how_to_change_backend.md)

2
examples/vision/keypointdetection/tiny_pose/python/README.md
View File

@@ -76,4 +76,4 @@ PP-TinyPose模型加载和初始化，其中model_file, params_file以及config_
- [PP-TinyPose 模型介绍](..)
- [PP-TinyPose C++部署](../cpp)
- [模型预测结果说明](../../../../../docs/api/vision_results/)
- [如何切换模型推理后端引擎](../../../../../docs/runtime/how_to_change_backend.md)
- [如何切换模型推理后端引擎](../../../../../docs/cn/faq/how_to_change_backend.md)

2
examples/vision/matting/ppmatting/cpp/README.md
View File

@@ -7,7 +7,7 @@
- 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. 根据开发环境，下载预编译部署库和samples代码，参考[FastDeploy预编译库](../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md)
以Linux上 PP-Matting 推理为例，在本目录执行如下命令即可完成编译测试（如若只需在CPU上部署，可在[Fastdeploy C++预编译库](../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md/CPP_prebuilt_libraries.md)下载CPU推理库）
以Linux上 PP-Matting 推理为例，在本目录执行如下命令即可完成编译测试（如若只需在CPU上部署，可在[Fastdeploy C++预编译库](../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md)下载CPU推理库）
```bash
#下载SDK，编译模型examples代码（SDK中包含了examples代码）

2
examples/vision/matting/rvm/cpp/README.md
View File

@@ -5,7 +5,7 @@
- 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. 根据开发环境，下载预编译部署库和samples代码，参考[FastDeploy预编译库](../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md)
以Linux上 RobustVideoMatting 推理为例，在本目录执行如下命令即可完成编译测试（如若只需在CPU上部署，可在[Fastdeploy C++预编译库](../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md/CPP_prebuilt_libraries.md)下载CPU推理库）
以Linux上 RobustVideoMatting 推理为例，在本目录执行如下命令即可完成编译测试（如若只需在CPU上部署，可在[Fastdeploy C++预编译库](../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md)下载CPU推理库）
本目录下提供`infer.cc`快速完成RobustVideoMatting在CPU/GPU，以及GPU上通过TensorRT加速部署的示例。执行如下脚本即可完成

5
examples/vision/ocr/PP-OCRv3/mini_program/README.md
View File

@@ -16,7 +16,7 @@ import * as ocr from "@paddle-js-models/ocr";
await ocr.init(detConfig, recConfig);
const res = await ocr.recognize(img, option, postConfig);
```
ocr模型加载和初始化，其中模型为Paddle.js模型格式，js模型转换方式参考[文档](../../../../application/web_demo/README.md)
ocr模型加载和初始化，其中模型为Paddle.js模型格式，js模型转换方式参考[文档](../../../../application/js/web_demo/README.md)
**init函数参数**
@@ -37,5 +37,4 @@ ocr模型加载和初始化，其中模型为Paddle.js模型格式，js模型转
- [PP-OCRv3 C++部署](../cpp)
- [模型预测结果说明](../../../../../docs/api/vision_results/)
- [如何切换模型推理后端引擎](../../../../../docs/cn/faq/how_to_change_backend.md)
- [PP-OCRv3模型web demo文档](../../../../application/web_demo/README.md)
- [PP-OCRv3模型web demo文档](../../../../application/js/web_demo/README.md)

2
examples/vision/ocr/PP-OCRv3/web/README.md
View File

@@ -16,7 +16,7 @@ import * as ocr from "@paddle-js-models/ocr";
await ocr.init(detConfig, recConfig);
const res = await ocr.recognize(img, option, postConfig);
```
ocr模型加载和初始化，其中模型为Paddle.js模型格式，js模型转换方式参考[文档](../../../../application/web_demo/README.md)
ocr模型加载和初始化，其中模型为Paddle.js模型格式，js模型转换方式参考[文档](../../../../application/js/web_demo/README.md)
**init函数参数**

2
examples/vision/segmentation/paddleseg/quantize/cpp/README.md
View File

@@ -8,7 +8,7 @@
### 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.)
- 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.

2
examples/vision/segmentation/paddleseg/quantize/python/README.md
View File

@@ -8,7 +8,7 @@
### 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.)
- 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

2
examples/vision/segmentation/paddleseg/rknpu2/python/README.md
View File

@@ -4,7 +4,7 @@
- 1. 软硬件环境满足要求，参考[FastDeploy环境要求](../../../../../../docs/cn/build_and_install/rknpu2.md)
【注意】如你部署的为**PP-Matting**、**PP-HumanMatting**以及**ModNet**请参考[Matting模型部署](../../../matting)
【注意】如你部署的为**PP-Matting**、**PP-HumanMatting**以及**ModNet**请参考[Matting模型部署](../../../../matting/)
本目录下提供`infer.py`快速完成PPHumanseg在RKNPU上部署的示例。执行如下脚本即可完成

6
examples/vision/segmentation/paddleseg/web/README.md
View File

@@ -7,7 +7,7 @@
## 前端部署PP-Humanseg v1模型
PP-Humanseg v1模型web demo部署及使用参考[文档](../../../../application/web_demo/README.md)
PP-Humanseg v1模型web demo部署及使用参考[文档](../../../../application/js/web_demo/README.md)
## PP-Humanseg v1 js接口
@@ -41,7 +41,3 @@ humanSeg.blurBackground(res)
**drawHumanSeg()函数参数**
> * **seg_values**(number[]): 输入参数，一般是getGrayValue函数计算的结果作为输入

2
examples/vision/tracking/pptracking/cpp/README.md
View File

@@ -7,7 +7,7 @@
- 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. 根据开发环境，下载预编译部署库和samples代码，参考[FastDeploy预编译库](../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md)
以Linux上 PP-Tracking 推理为例，在本目录执行如下命令即可完成编译测试（如若只需在CPU上部署，可在[Fastdeploy C++预编译库](../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md/CPP_prebuilt_libraries.md)下载CPU推理库）
以Linux上 PP-Tracking 推理为例，在本目录执行如下命令即可完成编译测试（如若只需在CPU上部署，可在[Fastdeploy C++预编译库](../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md)下载CPU推理库）
```bash
#下载SDK，编译模型examples代码（SDK中包含了examples代码）

14
fastdeploy/vision/common/processors/normalize.cc
View File

@@ -19,7 +19,7 @@ namespace vision {
Normalize::Normalize(const std::vector<float>& mean,
const std::vector<float>& std, bool is_scale,
const std::vector<float>& min,
const std::vector<float>& max) {
const std::vector<float>& max, bool swap_rb) {
FDASSERT(mean.size() == std.size(),
"Normalize: requires the size of mean equal to the size of std.");
std::vector<double> mean_(mean.begin(), mean.end());
@@ -50,6 +50,7 @@ Normalize::Normalize(const std::vector<float>& mean,
alpha_.push_back(alpha);
beta_.push_back(beta);
}
swap_rb_ = swap_rb;
}
bool Normalize::ImplByOpenCV(Mat* mat) {
@@ -57,6 +58,7 @@ bool Normalize::ImplByOpenCV(Mat* mat) {
std::vector<cv::Mat> split_im;
cv::split(*im, split_im);
if (swap_rb_) std::swap(split_im[0], split_im[2]);
for (int c = 0; c < im->channels(); c++) {
split_im[c].convertTo(split_im[c], CV_32FC1, alpha_[c], beta_[c]);
}
@@ -79,9 +81,13 @@ bool Normalize::ImplByFlyCV(Mat* mat) {
std[i] = 1.0 / alpha_[i];
mean[i] = -1 * beta_[i] * std[i];
}
std::vector<uint32_t> channel_reorder_index = {0, 1, 2};
if (swap_rb_) std::swap(channel_reorder_index[0], channel_reorder_index[2]);
fcv::Mat new_im(im->width(), im->height(),
fcv::FCVImageType::PKG_BGR_F32);
fcv::normalize_to_submean_to_reorder(*im, mean, std, std::vector<uint32_t>(),
fcv::normalize_to_submean_to_reorder(*im, mean, std, channel_reorder_index,
new_im, true);
mat->SetMat(new_im);
return true;
@@ -91,8 +97,8 @@ bool Normalize::ImplByFlyCV(Mat* mat) {
bool Normalize::Run(Mat* mat, const std::vector<float>& mean,
const std::vector<float>& std, bool is_scale,
const std::vector<float>& min,
const std::vector<float>& max, ProcLib lib) {
auto n = Normalize(mean, std, is_scale, min, max);
const std::vector<float>& max, ProcLib lib, bool swap_rb) {
auto n = Normalize(mean, std, is_scale, min, max, swap_rb);
return n(mat, lib);
}

14
fastdeploy/vision/common/processors/normalize.h
View File

@@ -23,7 +23,8 @@ class FASTDEPLOY_DECL Normalize : public Processor {
Normalize(const std::vector<float>& mean, const std::vector<float>& std,
bool is_scale = true,
const std::vector<float>& min = std::vector<float>(),
const std::vector<float>& max = std::vector<float>());
const std::vector<float>& max = std::vector<float>(),
bool swap_rb = false);
bool ImplByOpenCV(Mat* mat);
#ifdef ENABLE_FLYCV
bool ImplByFlyCV(Mat* mat);
@@ -44,14 +45,23 @@ class FASTDEPLOY_DECL Normalize : public Processor {
const std::vector<float>& std, bool is_scale = true,
const std::vector<float>& min = std::vector<float>(),
const std::vector<float>& max = std::vector<float>(),
ProcLib lib = ProcLib::DEFAULT);
ProcLib lib = ProcLib::DEFAULT, bool swap_rb = false);
std::vector<float> GetAlpha() const { return alpha_; }
std::vector<float> GetBeta() const { return beta_; }
bool GetSwapRB() {
return swap_rb_;
}
void SetSwapRB(bool swap_rb) {
swap_rb_ = swap_rb;
}
private:
std::vector<float> alpha_;
std::vector<float> beta_;
bool swap_rb_;
};
} // namespace vision
} // namespace fastdeploy

16
fastdeploy/vision/common/processors/normalize_and_permute.cc
View File

@@ -21,7 +21,8 @@ NormalizeAndPermute::NormalizeAndPermute(const std::vector<float>& mean,
const std::vector<float>& std,
bool is_scale,
const std::vector<float>& min,
const std::vector<float>& max) {
const std::vector<float>& max,
bool swap_rb) {
FDASSERT(mean.size() == std.size(),
"Normalize: requires the size of mean equal to the size of std.");
std::vector<double> mean_(mean.begin(), mean.end());
@@ -52,6 +53,7 @@ NormalizeAndPermute::NormalizeAndPermute(const std::vector<float>& mean,
alpha_.push_back(alpha);
beta_.push_back(beta);
}
swap_rb_ = swap_rb;
}
bool NormalizeAndPermute::ImplByOpenCV(Mat* mat) {
@@ -60,6 +62,7 @@ bool NormalizeAndPermute::ImplByOpenCV(Mat* mat) {
int origin_h = im->rows;
std::vector<cv::Mat> split_im;
cv::split(*im, split_im);
if (swap_rb_) std::swap(split_im[0], split_im[2]);
for (int c = 0; c < im->channels(); c++) {
split_im[c].convertTo(split_im[c], CV_32FC1, alpha_[c], beta_[c]);
}
@@ -94,8 +97,12 @@ bool NormalizeAndPermute::ImplByFlyCV(Mat* mat) {
std[i] = 1.0 / alpha_[i];
mean[i] = -1 * beta_[i] * std[i];
}
std::vector<uint32_t> channel_reorder_index = {0, 1, 2};
if (swap_rb_) std::swap(channel_reorder_index[0], channel_reorder_index[2]);
fcv::Mat new_im;
fcv::normalize_to_submean_to_reorder(*im, mean, std, std::vector<uint32_t>(),
fcv::normalize_to_submean_to_reorder(*im, mean, std, channel_reorder_index,
new_im, false);
mat->SetMat(new_im);
mat->layout = Layout::CHW;
@@ -106,8 +113,9 @@ bool NormalizeAndPermute::ImplByFlyCV(Mat* mat) {
bool NormalizeAndPermute::Run(Mat* mat, const std::vector<float>& mean,
const std::vector<float>& std, bool is_scale,
const std::vector<float>& min,
const std::vector<float>& max, ProcLib lib) {
auto n = NormalizeAndPermute(mean, std, is_scale, min, max);
const std::vector<float>& max, ProcLib lib,
bool swap_rb) {
auto n = NormalizeAndPermute(mean, std, is_scale, min, max, swap_rb);
return n(mat, lib);
}

14
fastdeploy/vision/common/processors/normalize_and_permute.h
View File

@@ -23,7 +23,8 @@ class FASTDEPLOY_DECL NormalizeAndPermute : public Processor {
NormalizeAndPermute(const std::vector<float>& mean,
const std::vector<float>& std, bool is_scale = true,
const std::vector<float>& min = std::vector<float>(),
const std::vector<float>& max = std::vector<float>());
const std::vector<float>& max = std::vector<float>(),
bool swap_rb = false);
bool ImplByOpenCV(Mat* mat);
#ifdef ENABLE_FLYCV
bool ImplByFlyCV(Mat* mat);
@@ -44,7 +45,7 @@ class FASTDEPLOY_DECL NormalizeAndPermute : public Processor {
const std::vector<float>& std, bool is_scale = true,
const std::vector<float>& min = std::vector<float>(),
const std::vector<float>& max = std::vector<float>(),
ProcLib lib = ProcLib::DEFAULT);
ProcLib lib = ProcLib::DEFAULT, bool swap_rb = false);
void SetAlpha(const std::vector<float>& alpha) {
alpha_.clear();
@@ -58,9 +59,18 @@ class FASTDEPLOY_DECL NormalizeAndPermute : public Processor {
beta_.assign(beta.begin(), beta.end());
}
bool GetSwapRB() {
return swap_rb_;
}
void SetSwapRB(bool swap_rb) {
swap_rb_ = swap_rb;
}
private:
std::vector<float> alpha_;
std::vector<float> beta_;
bool swap_rb_;
};
} // namespace vision
} // namespace fastdeploy

67
fastdeploy/vision/common/processors/transform.cc
View File

@@ -95,10 +95,77 @@ void FuseNormalizeHWC2CHW(
<< std::endl;
}
void FuseNormalizeColorConvert(
std::vector<std::shared_ptr<Processor>>* processors) {
// Fuse Normalize and BGR2RGB/RGB2BGR
int normalize_index = -1;
int color_convert_index = -1;
// If these middle processors are after BGR2RGB/RGB2BGR and before Normalize,
// we can still fuse Normalize and BGR2RGB/RGB2BGR
static std::unordered_set<std::string> middle_processors(
{"Resize", "ResizeByShort", "ResizeByLong", "Crop", "CenterCrop",
"LimitByStride", "LimitShort", "Pad", "PadToSize", "StridePad",
"WarpAffine"});
for (size_t i = 0; i < processors->size(); ++i) {
if ((*processors)[i]->Name() == "BGR2RGB" ||
(*processors)[i]->Name() == "RGB2BGR") {
color_convert_index = i;
for (size_t j = color_convert_index + 1; j < processors->size(); ++j) {
if ((*processors)[j]->Name() == "Normalize" ||
(*processors)[j]->Name() == "NormalizeAndPermute") {
normalize_index = j;
break;
}
}
if (normalize_index < 0) {
return;
}
for (size_t j = color_convert_index + 1; j < normalize_index; ++j) {
if (middle_processors.count((*processors)[j]->Name())) {
continue;
}
return;
}
}
}
if (color_convert_index < 0) {
return;
}
// Delete Color Space Convert
std::string color_processor_name = (*processors)[color_convert_index]->Name();
processors->erase(processors->begin() + color_convert_index);
// Toggle the swap_rb option of the Normalize processor
std::string normalize_processor_name =
(*processors)[normalize_index - 1]->Name();
bool swap_rb;
if (normalize_processor_name == "Normalize") {
auto processor = dynamic_cast<Normalize*>(
(*processors)[normalize_index - 1].get());
swap_rb = processor->GetSwapRB();
processor->SetSwapRB(!swap_rb);
} else if (normalize_processor_name == "NormalizeAndPermute") {
auto processor = dynamic_cast<NormalizeAndPermute*>(
(*processors)[normalize_index - 1].get());
swap_rb = processor->GetSwapRB();
processor->SetSwapRB(!swap_rb);
} else {
FDASSERT(false, "Something wrong in FuseNormalizeColorConvert().");
}
FDINFO << color_processor_name << " and " << normalize_processor_name
<< " are fused to " << normalize_processor_name
<< " with swap_rb=" << !swap_rb << std::endl;
}
void FuseTransforms(
std::vector<std::shared_ptr<Processor>>* processors) {
FuseNormalizeCast(processors);
FuseNormalizeHWC2CHW(processors);
FuseNormalizeColorConvert(processors);
}

4
fastdeploy/vision/common/processors/transform.h
View File

@@ -31,6 +31,7 @@
#include "fastdeploy/vision/common/processors/resize_by_short.h"
#include "fastdeploy/vision/common/processors/stride_pad.h"
#include "fastdeploy/vision/common/processors/warp_affine.h"
#include <unordered_set>
namespace fastdeploy {
namespace vision {
@@ -41,6 +42,9 @@ void FuseTransforms(std::vector<std::shared_ptr<Processor>>* processors);
void FuseNormalizeCast(std::vector<std::shared_ptr<Processor>>* processors);
// Fuse Normalize + HWC2CHW to NormalizeAndPermute
void FuseNormalizeHWC2CHW(std::vector<std::shared_ptr<Processor>>* processors);
// Fuse Normalize + Color Convert
void FuseNormalizeColorConvert(
std::vector<std::shared_ptr<Processor>>* processors);
} // namespace vision
} // namespace fastdeploy

4
java/android/app/src/main/java/com/baidu/paddle/fastdeploy/app/ui/view/ResultListView.java
View File

@@ -5,10 +5,6 @@ import android.os.Handler;
import android.util.AttributeSet;
import android.widget.ListView;
/**
* Created by ruanshimin on 2018/5/14.
*/
public class ResultListView extends ListView {
public ResultListView(Context context) {
super(context);

4
java/android/app/src/main/java/com/baidu/paddle/fastdeploy/app/ui/view/adapter/DetectResultAdapter.java
View File

@@ -15,10 +15,6 @@ import com.baidu.paddle.fastdeploy.app.ui.view.model.BaseResultModel;
import java.text.DecimalFormat;
import java.util.List;
/**
* Created by ruanshimin on 2018/5/13.
*/
public class DetectResultAdapter extends ArrayAdapter<BaseResultModel> {
private int resourceId;

4
java/android/app/src/main/java/com/baidu/paddle/fastdeploy/app/ui/view/model/BaseResultModel.java
View File

@@ -1,9 +1,5 @@
package com.baidu.paddle.fastdeploy.app.ui.view.model;
/**
* Created by ruanshimin on 2018/5/16.
*/
public class BaseResultModel {
private int index;
private String name;

4
java/android/app/src/main/res/values/strings.xml
View File

@@ -2,8 +2,8 @@
<!-- Default App name -->
<string name="app_name">EasyEdge</string>
<!-- Other App name -->
<string name="detection_app_name">FastDeploy PicoDet</string>
<string name="ocr_app_name">FastDeploy PP-OCRv2</string>
<string name="detection_app_name">EasyEdge</string>
<string name="ocr_app_name">EasyEdge</string>
<!-- Keys for PreferenceScreen -->
<string name="CHOOSE_PRE_INSTALLED_MODEL_KEY">CHOOSE_INSTALLED_MODEL_KEY</string>
<string name="MODEL_DIR_KEY">MODEL_DIR_KEY</string>

8
tests/models/test_mobilenetv2.py
View File

@@ -48,12 +48,10 @@ def test_classification_mobilenetv2():
im1 = cv2.imread("./resources/ILSVRC2012_val_00000010.jpeg")
im2 = cv2.imread("./resources/ILSVRC2012_val_00030010.jpeg")
# for i in range(3000000):
while True:
for i in range(3):
# test single predict
model.postprocessor.topk = 6
result1 = model.predict(im1)
result2 = model.predict(im2)
result1 = model.predict(im1, 6)
result2 = model.predict(im2, 6)
diff_label_1 = np.fabs(
np.array(result1.label_ids) - np.array(expected_label_ids_1))