HivisionIDPhoto 旨在开发一种实用、系统性的证件照智能制作算法。HivisionIDPhoto_cpp是对HivisionIDPhoto C++的重构,目的利用边缘设备的计算资源,以实现嵌入式设备和手机端本地部署。
HivisionIDPhoto_cpp可以做到。
1.离线手机端部署。arm设备部署运行。
2.轻量级抠图
3.根据不同尺寸规格生成不同的标准证件照
4.六寸排版照
5.美颜
6.安卓apk(waitting)
windows版和Linux版程序压缩放在release中,解压后需将执行文件和依赖环境放入HivisionIDPhotos_cpp根目录下,权重文件放入model文件夹下。
存到项目的model
目录下
modnet_photographic_portrait_matting.mnn 下载
hivision_modnet.mnn 下载
symbol_10_320_20L_5scales_v2_deploy.mnn下载
symbol_10_320_20L_8scales_v2_deploy.mnn下载
https://github.com/zjkhahah/HivisionIDPhotos_cpp.git
cd HivisionIDPhotos_cpp
其中MNN版本为2.9.0
opencv版本为4.7.0
- 环境要求
Microsoft Visual Studio >= 2017
cmake >= 3.13
powershell
Ninja
编译MNN库并把编译后的.a静态库文件放入lib文件夹下
编译MNN windows版教程为主库编译 — MNN-Doc 2.1.1 documentation
编译opencv windows版.a静态库文件放入lib文件夹下
- 编译代码
在设置中找到vcvars64.bat(适用于 VS 2017 的 x64 本机工具命令提示)并单击,打开VS编译x64架构程序的虚拟环境
cd HivisionIDPhotos_cpp
mkdir build && cd build
cmake .. -G Ninja #命令行编译cmake .. -G Ninja api编译 cmake .. -G Ninja -DCOMPILE_LIBRARY=ON
ninja
以Linaro工具链为例。首先从Linaro网页中按照宿主机以及交叉编译目标设备来选择合适的工具链。这里我们以arm-linux-gnueabi
为例,点击网页上的链接,进入arm-linux-gnueabi页面。 按照宿主机类型(这里以X64 Linux为例)选择下载链接, 文件名形如 gcc-linaro-7.5.0-2019.12-x86_64_arm-linux-gnueabi.tar.xz 下载后解压到任意目录
编译arm64版本的opencv和mnn
编译MNN ARM版教程为点击这里
编译好的静态opencv文件和mnn文件放入lib文件夹下
cd HivisionIDPhotos_cpp
mkdir build && cd build
cmake .. \
-DCMAKE_SYSTEM_NAME=宿主系统,例如Linux \
-DCMAKE_SYSTEM_VERSION=1 \
-DCMAKE_SYSTEM_PROCESSOR=交叉编译目标处理器的信息。例如aarch64 \
-DCMAKE_C_COMPILER=交叉编译器中C编译器的路径 \
-DCMAKE_CXX_COMPILER=交叉编译器中C++编译器的路径
make -j8
- 编译MNN 安卓版教程为[主库编译 — MNN-Doc 2.1.1 documentation (mnn-docs.readthedocs.io)]
- 下载opencv安卓版本链接
编译好的静态opencv文件和mnn文件放入lib文件夹下
cd HivisionIDPhotos_cpp
mkdir build && cd build
cmake .. -DCMAKE_TOOLCHAIN_FILE=#CMake工具链文件的路径 -DCMAKE_ANDROID_NDK=#Android NDK(Native Development Kit)的安装路径 -DANDROID_PLATFORM=android-21 -DANDROID_NATIVE_API_LEVEL=21 -DANDROID_ABI=arm64-v8a -DANDROID_NATIVE_API_LEVEL=21 -DCOMPILE_LIBRARY=ON -DCOMPILE_ANDROID=ON
make -j8
核心指令:
-e
: 人像在图片比例-i
:输入照片路径-o
:输出照片路径-k
:输出照片大小-f
:人脸模型选择- ...
更多指令通过./HivisionIDPhotos_cpp.exe --help
查看
以windows系统为例,使用工具powershell跳转到根目录下,例:
cd D:\HivisionIDPhotos-cpp
输入一张照片,输出一张标准证件照png
./HivisionIDPhotos_cpp.exe -i demo/images/test.jpg -o 1 -r 255 -g 0 -b 0 -h 413 -w 295
输入 1 张照片,获得 1张 4 通道透明 png
输入一张照片,得到一张六寸排版照 png
./HivisionIDPhotos_cpp.exe -i demo/images/test.jpg -o 1 -r 255 -g 0 -b 0 -h 413 -w 295 -l 1
利用SpringBoot+JNA调用dll文件生成API(linux环境调用.so文件) 下面以HivisionIDphotos中的human_mating方法为例
<dependency>
<groupId>net.java.dev.jna</groupId>
<artifactId>jna</artifactId>
<version>5.12.1</version>
<scope>compile</scope>
</dependency>
<dependency>
<groupId>org.projectlombok</groupId>
<artifactId>lombok</artifactId>
<version>1.18.8</version>
</dependency>
<dependency>
<groupId>com.alibaba</groupId>
<artifactId>fastjson</artifactId>
<version>1.2.28</version>
</dependency>
<dependency>
<groupId>commons-beanutils</groupId>
<artifactId>commons-beanutils</artifactId>
<version>1.9.2</version>
</dependency>
<dependency>
<groupId>commons-collections</groupId>
<artifactId>commons-collections</artifactId>
<version>3.2.1</version>
</dependency>
<dependency>
<groupId>commons-lang</groupId>
<artifactId>commons-lang</artifactId>
<version>2.6</version>
</dependency>
<dependency>
<groupId>commons-logging</groupId>
<artifactId>commons-logging</artifactId>
<version>1.1.1</version>
</dependency>
<dependency>
<groupId>net.sf.ezmorph</groupId>
<artifactId>ezmorph</artifactId>
<version>1.0.6</version>
</dependency>
将需要调用的dll文件放在src\main\resources\win32-x86-64\中
这里需要知道方法的具体传参,根据JNA映射规则进行编写,下面三个类根据HivisionIDphotos.dll内函数的传参结构体编写。 Hivision_java_params.java
public class Hivision_java_params extends Structure {
public String model_path;
public String image_path;
public String out_path;
public String face_model_path;
public int rgb_r, rgb_g, rgb_b;
public int thread_num, model_scale;
public Params param = new Params();
@Override
protected List<String> getFieldOrder() {
return Arrays.asList("model_path", "image_path", "out_path", "face_model_path",
"rgb_r", "rgb_g", "rgb_b", "thread_num", "model_scale", "param");
}
public Hivision_java_params() {
super();
// 初始化默认值
this.rgb_r = 255;
this.rgb_g = 0;
this.rgb_b = 0;
this.thread_num = 4;
this.model_scale = 8;
}
}
Params.java
public class Params extends Structure {
public int out_image_width, out_image_height;
public boolean change_bg_only;
public float head_measure_ratio, head_height_ratio;
public float[] head_top_range = new float[2];
public int rgb_r, rgb_g, rgb_b;
public FaceInfo face_info = new FaceInfo();
@Override
protected List<String> getFieldOrder() {
return Arrays.asList("out_image_width", "out_image_height", "change_bg_only",
"head_measure_ratio", "head_height_ratio", "head_top_range",
"rgb_r", "rgb_g", "rgb_b", "face_info");
}
public Params() {
super();
// 初始化默认值
out_image_width = 295;
out_image_height = 413;
change_bg_only = false;
head_measure_ratio = 0.2f;
head_height_ratio = 0.55f;
head_top_range[0] = 0.12f;
head_top_range[1] = 0.1f;
rgb_r = 255;
rgb_g = 0;
rgb_b = 0;
}}
FaceInfo.java
public class FaceInfo extends Structure {
public float x1, y1, x2, y2, score, area;
public float[] landmarks = new float[10];
@Override
protected List<String> getFieldOrder() {
return Arrays.asList("x1", "y1", "x2", "y2", "score", "area", "landmarks");
}
public FaceInfo() {
super();
// 初始化数组
Arrays.fill(landmarks, 0.0f);
}}
一个Library接口对应一个dll,内抽象方法与dll中的函数一一对应。Native.loadLibrary()第一个参数为dll文件路径,不加.dll后缀。 HivisionIDphotosLibrary.java
public interface HivisionIDphotosLibrary extends Library {
HivisionIDphotosLibrary INSTANCE = Native.loadLibrary("src\\main\\resources\\win32-x86-64\\HivisionIDphotos", HivisionIDphotosLibrary.class);
void human_mating(Hivision_java_params hivision_java_params);
int ID_photo(Hivision_java_params hivision_java_params,int out_size_kb,boolean layout_phot);}
JnaDemoController.java
@RestController
@RequestMapping("/JNA")
public class JnaDemoController {
@PostMapping("/human_mating")
public String human_mating(@RequestParam("hivision_java_params_json") String hivision_java_params_json){
try{
Hivision_java_params hivision_java_params = JSONObject.parseObject(hivision_java_params_json, Hivision_java_params.class);
HivisionIDphotosLibrary.INSTANCE.human_mating(hivision_java_params);
return "success";
}catch (Exception e){
System.out.println(e.getMessage());
return "fail";
}
}}
- MNN:
@software{Alibaba_MNN_2024,
author = {Alibaba},
title = {{MNN}},
url = {https://github.com/alibaba/MNN},
year = {2024},
publisher = {GitHub}
}
@software{Zeyi-Lin_HivisionIDPhotos_2024,
author = {Zeyi-Lin},
title = {{HivisionIDPhotos}},
url = {https://github.com/Zeyi-Lin/HivisionIDPhotos},
year = {2024},
publisher = {GitHub}
}
@software{LFFD-MNN_2024,
author = {SyGoing},
title = {{LFFD-MNN}},
url = {https://github.com/SyGoing/LFFD-MNN},
year = {2024},
publisher = {GitHub}
}
如果您有任何问题,请提issue或发邮件至junkangzhou@stu.xidian.edu.cn