libFaceSDK.so这个文件，能否改个名字？目前与百度的活体检测库存在so文件命名冲突问题，导致无法导入项目

因為想要拿Android舊手機當作"類監視器"

當人體移動進入鏡頭範圍(出 入)

就 拍照/錄影

謝謝

Have any plan to support super resolution with TengineKit?

请问人脸关键点检测模型，支持近红外图像或者红外图像吗？

如果直接编译 Linux 版本，会出现和 #15 一样的问题。

库 libTengineKitSDK.so 中，tenginekit_api.h 引用的 facesdk_readModelFromFile 函数的参数和 demo.cpp 中的函数参数不一样，多了一个 bool 参数，我也不知道这个参数的具体意义。

自己在 tenginekit_api.h 加入了这个 bool 参数，比如将

void facesdk_readModelFromFile(ModelType type, const char* model_path, ImageFormat modelInputFormat);

改为

void facesdk_readModelFromFile(ModelType type, const char* model_path, ImageFormat modelInputFormat, bool why);

之后，能够编译成功。但运行识别会报错：

terminate called after throwing an instance of 'std::bad_alloc'
  what():  std::bad_alloc

colab 复现链接 可以复现这个报错

我看了一下，是 demo.app 的第 45 行会报错：

facesdk_readModelFromFile(ModelType::Detect, "models/detect.bin", ImageFormat::RGB, true);

另外，如果把 45 行的最后那个 bool 参数改为 false 的话：

facesdk_readModelFromFile(ModelType::Detect, "models/detect.bin", ImageFormat::RGB, false);

45 行不会报错，但 46 行会报错：

sdkFaces faces = facesdk_detect(data);

这种情况下，colab 不会显示错误，会直接退出运行，但在我自己机子上会显示这个错误：

Segmentation fault (core dumped)

目前用的工控板很多都是RK3288或者RK3399，这个sdk在RK3399主板上CPU使用率到达了20%+，然后加上UI绘制，CPU的使用率就比较高了，对比阿里开源的人脸检测SDK，阿里的CPU使用率在RK3399主板上一般是5%左右，希望开源对CPU使用率进行优化

Hello,
Now I want to get the effect like your example in my own image by using TengineKit. The program can generate eye_landmarks and eye_iris, and they all in 3D coordinate, but I have no idea to convert those points to 2D coordinate and draw them on my own image. Should I need to calibrate my camera, so the 3D coordinates points can map to 2D coordinate? Could your please tell me some details about it or give me a demo.
Thanks

"由于jfrog将于2021年5月1日后无法更新，我们无法持续更新SDK库。同时jfrog即将于2022年2月1日停止提供服务，我们的SDK也将在时间点后无法提供正常的服务。"

Khandas里的人脸模型在CPU上推理效果不好，同样的示例图片会出多个检测框

代码如下
KitCore.init(context, AndroidConfig
.create()
.setNormalMode()
.openFunc(AndroidConfig.Func.Detect)
.openFunc(AndroidConfig.Func.Attribution)
.setInputImageFormat(AndroidConfig.ImageFormat.RGB))；
val detect = Face.detect(bitmap2RGBData(bmp))

或者每次都初始化
KitCore.init(context, AndroidConfig
.create()
.setNormalMode()
.openFunc(AndroidConfig.Func.Detect)
.openFunc(AndroidConfig.Func.Attribution)
.setInputImageFormat(AndroidConfig.ImageFormat.RGB)
.setInputImageSize(bmp.width,bmp.height)
.setOutputImageSize(bmp.width,bmp.height))
val detect = Face.detect(bitmap2RGBData(bmp))

最终返回的faceCount都是0

object THTengineKit: BaseScanCore() {
    init {
        KitCore.init(BaseApplication.application, AndroidConfig
            .create()
            .setDefaultFunc()
            .setNormalMode()
            .setInputImageFormat(AndroidConfig.ImageFormat.RGBA)
        )

        val state = KitCore.getInit()
    }

    override fun doAI(filePath: String): Int {
        val bmp = BitmapFactory.decodeFile(filePath)
        if(bmp != null && !bmp.isRecycled) {
            val faceDetect = Face.detect(getPixelsRGBA(bmp))
//            bmp.recycle()
            val faceDetectInfos = faceDetect.detectInfos
            // 这里一直返回0
            return faceDetectInfos.size
        }
        return 0
    }

    override fun release() {
    }
}

getInit()方法拿到了true，应该是初始化成功的。
getPixelsRGBA方法是closed issue里面的那个方法

fun getPixelsRGBA(image: Bitmap): ByteArray {
        // calculate how many bytes our image consists of
        val bytes = image.byteCount
        val buffer: ByteBuffer = ByteBuffer.allocate(bytes) // Create a new buffer
        image.copyPixelsToBuffer(buffer) // Move the byte data to the buffer
        return buffer.array() // Get the underlying array containing the
}

是哪里出了问题呢？？？感谢解答

Hi! Yout lib is very useful, I'd like to use it in production app, but have some suggestions/remarks:

1. BlazeFace function should have 6 landmarks besides bounding box, but if I try for single openFunc(AndroidConfig.Func.BlazeFace) make faceDetect.landmark2d(), it will crash.
2. The lib only have arm .so libs, but doesn't have x86 and x86_64, but google play request all architectures

Android CameraX 返回的数据是YUV_420_888 格式的

/zygote: Long monitor contention with owner demo-pool-3 (7295) at com.tenginekit.Face$FaceDetect com.tenginekit.Face.detect(byte[])(Face.java:-1) waiters=2 in com.tenginekit.Face$FaceDetect com.tenginekit.Face.detect(byte[]) for 169ms

linux 编译报错

我同样有在编译时错误
执行如下：sudo sh make.sh
[sudo] maphysart 的密码：
-- The C compiler identification is GNU 7.5.0
-- The CXX compiler identification is GNU 7.5.0
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Found OpenCV: /usr/local (found version "4.1.2")
-- Configuring done
-- Generating done
-- Build files have been written to: /home/maphysart/Desktop/Marks/TengineKit-test/Linux/sample/FaceDemo/build
Scanning dependencies of target FaceSDKDemo
[ 50%] Building CXX object CMakeFiles/FaceSDKDemo.dir/src/demo.cpp.o
In file included from /home/maphysart/Desktop/Marks/TengineKit-test/Linux/sample/FaceDemo/src/tenginekit_api.h:7:0,
from /home/maphysart/Desktop/Marks/TengineKit-test/Linux/sample/FaceDemo/src/demo.cpp:5:
/home/maphysart/Desktop/Marks/TengineKit-test/Linux/sample/FaceDemo/src/tenginekit_struct.h:10:1: warning: ‘typedef’ was ignored in this declaration
typedef struct vector2 {
^~~~~~~
/home/maphysart/Desktop/Marks/TengineKit-test/Linux/sample/FaceDemo/src/tenginekit_struct.h:52:1: warning: ‘typedef’ was ignored in this declaration
typedef struct BodyInfo
^~~~~~~
/home/maphysart/Desktop/Marks/TengineKit-test/Linux/sample/FaceDemo/src/tenginekit_struct.h:58:1: warning: ‘typedef’ was ignored in this declaration
typedef struct HandInfo3d
^~~~~~~
[100%] Linking CXX executable ../bin/FaceSDKDemo
/usr/bin/ld: warning: libcurl.so, needed by /home/maphysart/Desktop/Marks/TengineKit-test/Linux/sample/FaceDemo/libs/libTengineKitSDK.so, not found (try using -rpath or -rpath-link)
/home/maphysart/Desktop/Marks/TengineKit-test/Linux/sample/FaceDemo/libs/libTengineKitSDK.so：对‘curl_easy_perform’未定义的引用
/home/maphysart/Desktop/Marks/TengineKit-test/Linux/sample/FaceDemo/libs/libTengineKitSDK.so：对‘curl_easy_getinfo’未定义的引用
/home/maphysart/Desktop/Marks/TengineKit-test/Linux/sample/FaceDemo/libs/libTengineKitSDK.so：对‘curl_easy_init’未定义的引用
/home/maphysart/Desktop/Marks/TengineKit-test/Linux/sample/FaceDemo/libs/libTengineKitSDK.so：对‘curl_slist_append’未定义的引用
/home/maphysart/Desktop/Marks/TengineKit-test/Linux/sample/FaceDemo/libs/libTengineKitSDK.so：对‘curl_version_info’未定义的引用
/home/maphysart/Desktop/Marks/TengineKit-test/Linux/sample/FaceDemo/libs/libTengineKitSDK.so：对‘curl_easy_cleanup’未定义的引用
/home/maphysart/Desktop/Marks/TengineKit-test/Linux/sample/FaceDemo/libs/libTengineKitSDK.so：对‘curl_slist_free_all’未定义的引用
/home/maphysart/Desktop/Marks/TengineKit-test/Linux/sample/FaceDemo/libs/libTengineKitSDK.so：对‘curl_formfree’未定义的引用
/home/maphysart/Desktop/Marks/TengineKit-test/Linux/sample/FaceDemo/libs/libTengineKitSDK.so：对‘curl_formadd’未定义的引用
/home/maphysart/Desktop/Marks/TengineKit-test/Linux/sample/FaceDemo/libs/libTengineKitSDK.so：对‘curl_easy_setopt’未定义的引用
collect2: error: ld returned 1 exit status
CMakeFiles/FaceSDKDemo.dir/build.make:144: recipe for target '../bin/FaceSDKDemo' failed
make[2]: *** [../bin/FaceSDKDemo] Error 1
CMakeFiles/Makefile2:67: recipe for target 'CMakeFiles/FaceSDKDemo.dir/all' failed
make[1]: *** [CMakeFiles/FaceSDKDemo.dir/all] Error 2
Makefile:83: recipe for target 'all' failed
make: *** [all] Error 2

怎样解决？

Glad to see your successful outcomes.

I am doing a project related to face landmark detection, but we use python to realize.
Is there a python api for this project?
thanks

你好 后期会考虑开源训练源码

你好，这个SDK支持iphone吗

First of all, I really like the code and it works pretty fast. I downloaded the sample apk and install to my Android phone. It is super fast. However, when I compile the sample code, it has a little bit delay. Can you explain what kind of optimizations did you use, or if it is possible can you upload the source code of apk?

On the android samples for face, the variable isProcessingFrame on line 115 in the CameraActivity.java, located wrong. It must be set to false after processImage() on line 116. Because when the image processing, the OnPreviewFrame function must be locked. With that correction, fps rate gets a little lower, however, works properly. You can simply change the places of these two lines.

Hello, which landmark detect network is used on this apk?

没有检测结果的输出提示如下：
Tengine plugin allocator TIMVX is registered.
vsiplugin reshape output tensor addr is 0x55a4c31d10
vsiplugin reshape output tensor addr is 0x55a4c31db0
vsiplugin reshape output tensor addr is 0x55a4c31e50
vsiplugin reshape output tensor addr is 0x55a4c31ef0
vsiplugin reshape output tensor addr is 0x55a4c32530
vsiplugin reshape output tensor addr is 0x55a4c320b0
vsiplugin reshape output tensor addr is 0x55a4c32150
vsiplugin reshape output tensor addr is 0x55a4c321f0
TENGINE_LAYOUT_NCHW goto RELEASE
data size is 4472.
data size is 4472.
form : 3 ,to : 0
vsiplugin reshape output tensor addr is 0x55a5af9cc0
TENGINE_LAYOUT_NCHW goto RELEASE
data size is 128.
TENGINE_LAYOUT_NCHW goto RELEASE
data size is 1724.
init Over
Detect Run Graph cost:1.22803
1
Run Over
Tensor size mismatch 256 vs 512.
/home/khadas/projects/khadas/vsiplugin-master/vxnpu/driver/vx_device.c:363 set input data for tensor 0 failed
run subgraph 2 error!
Landmark Run Graph cost:1.41992
Run Over
############################
0，环境Ubuntu20.04，[ 1190.488197] Galcore version 6.4.3.p0.286725
1、采用最新的tengine编译的libtengine-lite.so，采用自带的tm_classification_timvx.c这个函数得到的可执行程序是可以使用npu的。
2、sudo apt install tengine-libs采用这种方式安装，解决了“load plugin failed: libvxplugin.so: cannot open shared object file: No such file or directory”

也支持ios吧

您好，非常感谢贵团队的工作。

我在使用手部检测 demo 的时候发现，当镜头中同时有两只手 A 和 B 时，只有两只手在镜头中的距离非常非常远的时候，才能在当前帧处理结果中拿到两只手的信息。如若不然，比如相对距离较近（但肯定没有手部遮挡和交叠），就只能拿到一只手的信息，在接下来的帧处理中可能会频繁交替地返回手 A 或者手 B 的信息（但总归只有一只手）。

我认为这一点不是很合理，是做了什么非极大值抑制吗？这一点可以更改优化吗？盼回～

facesdk_readModelFromFile(ModelType::Detect, "../models/palm_detection_builtin.bin", ImageFormat::RGB); //读取模型 std::cout << "handsdfd:" << std::endl; sdkHand3d hands = facesdk_handdetect(data);//手部检测
使用sdkHand3d hands = facesdk_handdetect(data);//手部检测但是会出现**Segmentation fault (core dumped)**错误？
请问这个如何解决

可以使用 MediaDrm 提供的这个id, 除了部分手机恢复出厂设置, 这个id都不变的.
UUID wideVineUuid = new UUID(-0x121074568629b532L, -0x5c37d8232ae2de13L);
wvDrm = new MediaDrm(wideVineUuid);
byte[] wideVineId = wvDrm.getPropertyByteArray(MediaDrm.PROPERTY_DEVICE_UNIQUE_ID);

鸿蒙手机，在检测时晃动俩下就会出现如下错误：

2021-12-02 15:14:38.595 11099-11099/? A/DEBUG: Timestamp: 2021-12-02 15:14:38+0800
2021-12-02 15:14:38.595 11099-11099/? A/DEBUG: pid: 32600, tid: 10911, name: Thread-12 >>> com.daofeng.shanghubao <<<
2021-12-02 15:14:38.595 11099-11099/? A/DEBUG: uid: 10808
2021-12-02 15:14:38.595 11099-11099/? A/DEBUG: signal 11 (SIGSEGV), code 1 (SEGV_MAPERR), fault addr 0xaff80000
2021-12-02 15:14:38.595 11099-11099/? A/DEBUG: r0 aff80000 r1 b05c2b24 r2 fff7ff7c r3 00000000
2021-12-02 15:14:38.595 11099-11099/? A/DEBUG: r4 b0542ae4 r5 00000008 r6 aff00000 r7 bb6f26a8
2021-12-02 15:14:38.595 11099-11099/? A/DEBUG: r8 b0500000 r9 00000008 r10 000001e0 r11 fffffffc
2021-12-02 15:14:38.595 11099-11099/? A/DEBUG: ip f35344ec sp bb6f2678 lr bcf9d6a7 pc f34d9048
2021-12-02 15:14:38.712 11099-11099/? A/DEBUG: backtrace:
2021-12-02 15:14:38.712 11099-11099/? A/DEBUG: #00 pc 0004f048 /apex/com.android.runtime/lib/bionic/libc.so (__memcpy_a15+112) (BuildId: b3a547af3eb880c5dfd66f61a8c8b4ca)
2021-12-02 15:14:38.712 11099-11099/? A/DEBUG: #1 pc 0003b6a3 /data/app/com.daofeng.shanghubao-3GQsNb2-hMt0ZcNIW4Qdcg==/lib/arm/libtenginekit_engine.so (ImageResizeHelper::cropYuv(unsigned char const*, unsigned char*, int, int, int, int, int, int)+52) (BuildId: a7eb36802f61e8b75f977c85772960cd9b20aaab)
2021-12-02 15:14:38.712 11099-11099/? A/DEBUG: #2 pc 0001217b /data/app/com.daofeng.shanghubao-3GQsNb2-hMt0ZcNIW4Qdcg==/lib/arm/libtenginekit_engine.so (faceService::runEyeLandmark(FaceInfo&, unsigned char const*, int, int, bool)+378) (BuildId: a7eb36802f61e8b75f977c85772960cd9b20aaab)
2021-12-02 15:14:38.712 11099-11099/? A/DEBUG: #3 pc 00011d81 /data/app/com.daofeng.shanghubao-3GQsNb2-hMt0ZcNIW4Qdcg==/lib/arm/libtenginekit_engine.so (faceService::runDetect(unsigned char const*, int, int, std::__ndk1::vector<FaceInfo, std::__ndk1::allocator>&, FaceConfig const&)+332) (BuildId: a7eb36802f61e8b75f977c85772960cd9b20aaab)
2021-12-02 15:14:38.712 11099-11099/? A/DEBUG: #4 pc 0000f077 /data/app/com.daofeng.shanghubao-3GQsNb2-hMt0ZcNIW4Qdcg==/lib/arm/libtenginekit_engine.so (manager::detectFace(unsigned char const*, std::__ndk1::vector<FaceInfo, std::__ndk1::allocator>&, int, int, FaceConfig const&)+22) (BuildId: a7eb36802f61e8b75f977c85772960cd9b20aaab)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #5 pc 0001b76f /data/app/com.daofeng.shanghubao-3GQsNb2-hMt0ZcNIW4Qdcg==/lib/arm/libtenginekit_engine.so (Java_com_tenginekit_engine_core_TengineKitEngine_nativeDetectFace+526) (BuildId: a7eb36802f61e8b75f977c85772960cd9b20aaab)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #6 pc 00071853 /data/app/com.daofeng.shanghubao-3GQsNb2-hMt0ZcNIW4Qdcg==/oat/arm/base.odex (art_jni_trampoline+122)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #7 pc 02009a45 /memfd:/jit-cache (deleted) (com.daofengfun.facevideowh.FFmpegRecordActivity.processImage+1604)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #8 pc 000e3bc5 /apex/com.android.runtime/lib/libart.so (art_quick_invoke_stub_internal+68) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #9 pc 004558bb /apex/com.android.runtime/lib/libart.so (art_quick_invoke_stub+250) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #10 pc 000ebffd /apex/com.android.runtime/lib/libart.so (art::ArtMethod::Invoke(art::Thread*, unsigned int*, unsigned int, art::JValue*, char const*)+160) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #11 pc 0022499f /apex/com.android.runtime/lib/libart.so (art::interpreter::ArtInterpreterToCompiledCodeBridge(art::Thread*, art::ArtMethod*, art::ShadowFrame*, unsigned short, art::JValue*)+274) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #12 pc 00220b07 /apex/com.android.runtime/lib/libart.so (bool art::interpreter::DoCall<false, false>(art::ArtMethod*, art::Thread*, art::ShadowFrame&, art::Instruction const*, unsigned short, art::JValue*)+802) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #13 pc 0044c96f /apex/com.android.runtime/lib/libart.so (MterpInvokeDirect+358) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #14 pc 000de914 /apex/com.android.runtime/lib/libart.so (mterp_op_invoke_direct+20) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #15 pc 003eaf1c /data/app/com.daofeng.shanghubao-3GQsNb2-hMt0ZcNIW4Qdcg==/oat/arm/base.vdex (com.daofengfun.facevideowh.FFmpegRecordActivity.access$1400)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #16 pc 0044d2f1 /apex/com.android.runtime/lib/libart.so (MterpInvokeStatic+932) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #17 pc 000de994 /apex/com.android.runtime/lib/libart.so (mterp_op_invoke_static+20) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #18 pc 003ea4ce /data/app/com.daofeng.shanghubao-3GQsNb2-hMt0ZcNIW4Qdcg==/oat/arm/base.vdex (com.daofengfun.facevideowh.FFmpegRecordActivity$ProcessImageThread.run+42)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #19 pc 00200bb3 /apex/com.android.runtime/lib/libart.so (_ZN3art11interpreterL7ExecuteEPNS_6ThreadERKNS_20CodeItemDataAccessorERNS_11ShadowFrameENS_6JValueEbb.llvm.14250088896784669869+266) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #20 pc 00205465 /apex/com.android.runtime/lib/libart.so (art::interpreter::EnterInterpreterFromEntryPoint(art::Thread*, art::CodeItemDataAccessor const&, art::ShadowFrame*)+120) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #21 pc 0043ed29 /apex/com.android.runtime/lib/libart.so (artQuickToInterpreterBridge+832) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #22 pc 000e85a1 /apex/com.android.runtime/lib/libart.so (art_quick_to_interpreter_bridge+32) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #23 pc 000e3bc5 /apex/com.android.runtime/lib/libart.so (art_quick_invoke_stub_internal+68) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #24 pc 004558bb /apex/com.android.runtime/lib/libart.so (art_quick_invoke_stub+250) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #25 pc 000ebffd /apex/com.android.runtime/lib/libart.so (art::ArtMethod::Invoke(art::Thread*, unsigned int*, unsigned int, art::JValue*, char const*)+160) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #26 pc 003932bb /apex/com.android.runtime/lib/libart.so (art::(anonymous namespace)::InvokeWithArgArray(art::ScopedObjectAccessAlreadyRunnable const&, art::ArtMethod*, art::(anonymous namespace)::ArgArray*, art::JValue*, char const*)+54) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #27 pc 00393fc5 /apex/com.android.runtime/lib/libart.so (art::InvokeVirtualOrInterfaceWithJValues(art::ScopedObjectAccessAlreadyRunnable const&, _jobject*, _jmethodID*, jvalue const*)+300) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #28 pc 003c7f0b /apex/com.android.runtime/lib/libart.so (art::Thread::CreateCallback(void*)+982) (BuildId: f19139b4309180c887ec6b2ca9af235c)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #29 pc 0009c5e7 /apex/com.android.runtime/lib/bionic/libc.so (__pthread_start(void*)+20) (BuildId: b3a547af3eb880c5dfd66f61a8c8b4ca)
2021-12-02 15:14:38.713 11099-11099/? A/DEBUG: #30 pc 00055a97 /apex/com.android.runtime/lib/bionic/libc.so (__start_thread+30) (BuildId: b3a547af3eb880c5dfd66f61a8c8b4ca)

insight下不同人脸feature特征值512维为啥都是一样，而且值特别奇怪

I'd like to contribute to this project, but I failed to setup devenv. Can you please provide more instructions?

Hi.
Follow your demo, I guess it should read the detect model at first, so
facesdk_readModelFromFile(ModelType::Detect, "../models/detect.bin", ImageFormat::RGB);
sdkFaces faces = facesdk_detect(data);

and then, I try to read the iris_landmark.bin model,
facesdk_readModelFromFile(ModelType::Iris, "../models/iris_landmark.bin", ImageFormat::RGB);
sdkFaces3d faces1 = facesdk_eyelandmark3d();

However, the faces1.face3d_count always equals to 0.
Anything wrong in my steps?

terminate called after throwing an instance of 'std::bad_alloc'
what(): std::bad_alloc
Aborted (core dumped)

Hello!
Thanks for your work, it is really nice.

Can you please provide information about the models: architectures, source training framework, training data, evaluation metrics? Maybe you have a paper describing any of these?

Thanks in advance.

Do you have any image like landmark points order?
I really need facemesh points order...

I changed camera index from 0 to 1. (back camera)
and the recognitions all fail.
any potential fix to this?

It only works well for front camera frames, not back camera frames.

如果我用yolox训练了一个行人检测模型，转成tengine格式后，怎么替换进去？有没有相关教程啥的？

貌似assets里面有不同功能的bin，有些功能的不需要阿

这个关键点的检测效果真心不错，很好奇怎么实现了几乎没有抖动的

Hello TengineKit,

Please adivce:
TengineKit/Khadas/sample/Face/bin/Face crash with error of "load plugin failed: libvxplugin.so: cannot open shared object file: No such file or directory"

Note:
A similar library (OpenVX implementation?) exist with the Khadas VIM3 NPU SDK at:
.../aml_npu_sdk_6.4.3/linux_sdk/linux_sdk_6.4.3/build/sdk/drivers_64/libovxlib.so

Appended below are the detailes of the build and run.

Build the VIM3 NPU test - ​https://github.com/OAID/TengineKit :

Reference:

• OpenVX - ​https://en.wikipedia.org/wiki/OpenVX

1. SSH to a Khadas VIM3 board (i.e. Amlogic A311D 4GB RAM with a 5 TOPS int8 NPU - ​​https://www.khadas.com/product-page/vim3 )
   ssh -X noveto@vim3-001
   1.1. The Khadas VIM3 board is installed with a Ubuntu 20.04 desktop and the default 4.9.x kernel:

$ uname -a
Linux vim3-001 4.9.241 #1 SMP PREEMPT Sun Dec 27 17:09:46 IST 2020 aarch64 aarch64 aarch64 GNU/Linux

$ lsb_release -a
No LSB modules are available.
Distributor ID: Ubuntu
Description: Ubuntu 20.04.1 LTS
Release: 20.04
Codename: focal

$ g++ --version
g++ (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0
Copyright (C) 2019 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

2. Setup dependencies
   sudo apt-get install cmake make g++ git

3. Build the TEngine lite dependency - follow build instructions from ​https://gitee.com/OAL/Tengine/blob/tengine-lite/doc/compile.md
   cd /opt/noveto/src/FaceDetection
   ... ... ...
   3.1. After build + install, the install located is /opt/noveto/src/FaceDetection/Tengine-Lite/build/install

4. Download sources of TengineKit
   cd /opt/noveto/src/FaceDetection
   git clone https://github.com/OAID/TengineKit && cd TengineKit

5. Verify
   $ git log
   commit 5672e38 (HEAD -> master, origin/master, origin/HEAD)
   Author: zhangjun [email protected]
   Date: Fri Jan 15 17:46:22 2021 +0800

   Linux sample
   ... ... ...

6. Fixup: add the expected link to the TEngine lite libraries
   cd /opt/noveto/src/FaceDetection/TengineKit/Khadas/sample/Face
   ln -s /opt/noveto/src/FaceDetection/Tengine-Lite/build/install/lib libs

7. Build
   cd /opt/noveto/src/FaceDetection/TengineKit/Khadas/sample/Face
   bash make.sh

8. Run
   cd /opt/noveto/src/FaceDetection/TengineKit/Khadas/sample/Face/bin
   ./Face
   load plugin failed: libvxplugin.so: cannot open shared object file: No such file or directory Load vx plugin failed.
   Segmentation fault (core dumped)
   8.1. Look for the offending code (i.e. the loading of libvxplugin.so)
   $ git blame Khadas/sample/Face/src/manager.cpp | grep libvxplugin.so
   b6cd822 (zhangjun 2021-01-04 15:23:04 +0800 19) if (load_tengine_plugin(VXDEVICE, "libvxplugin.so", "vx_plugin_init") < 0) {
   8.2. the change description is:
   $ git show b6cd822 | head
   commit b6cd822
   Author: zhangjun [email protected]
   Date: Mon Jan 4 15:23:04 2021 +0800

   update khadas