Android 端 arm64中libSeetaMaskDetector200.so调用出现错误
libSeetaMaskDetector200_java.so为自定义的jni接口调用libSeetaMaskDetector200.so
如果官方可以提供arm64、armv7的java层接口最好。
2020-05-06 17:23:13.631 17343-17343/? A/DEBUG: *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ***
2020-05-06 17:23:13.631 17343-17343/? A/DEBUG: Build fingerprint: 'OPPO/PBCM30/PBCM30:8.1.0/OPM1.171019.011/1584434156:user/release-keys'
2020-05-06 17:23:13.631 17343-17343/? A/DEBUG: Revision: '0'
2020-05-06 17:23:13.631 17343-17343/? A/DEBUG: ABI: 'arm64'
2020-05-06 17:23:13.631 17343-17343/? A/DEBUG: pid: 17235, tid: 17335, name: Thread-2 >>> com.seetatech.fas <<<
2020-05-06 17:23:13.631 17343-17343/? A/DEBUG: signal 6 (SIGABRT), code -6 (SI_TKILL), fault addr --------
2020-05-06 17:23:13.631 17343-17343/? A/DEBUG: x0 0000000000000000 x1 00000000000043b7 x2 0000000000000006 x3 0000000000000008
2020-05-06 17:23:13.631 17343-17343/? A/DEBUG: x4 fefefefefeff2d6b x5 fefefefefeff2d6b x6 fefefefefeff2d6b x7 7f7f7f7f7f7f7f7f
2020-05-06 17:23:13.631 17343-17343/? A/DEBUG: x8 0000000000000083 x9 0000000010000000 x10 0000007e7ff07050 x11 f89c91284a11cb20
2020-05-06 17:23:13.631 17343-17343/? A/DEBUG: x12 f89c91284a11cb20 x13 0000000000000020 x14 ffffffffffffffdf x15 00000000000043b7
2020-05-06 17:23:13.631 17343-17343/? A/DEBUG: x16 00000006b31adfa8 x17 0000007f1c47a540 x18 0000000000000000 x19 0000000000004353
2020-05-06 17:23:13.631 17343-17343/? A/DEBUG: x20 00000000000043b7 x21 0000000000000083 x22 0000007e90e33890 x23 0000007e7ff078a0
2020-05-06 17:23:13.631 17343-17343/? A/DEBUG: x24 0000007e7fadc020 x25 0000007e7f6b3fa0 x26 0000007e7ff08588 x27 0000007e7ff07580
2020-05-06 17:23:13.631 17343-17343/? A/DEBUG: x28 0000007e90e600b8 x29 0000007e7ff07090 x30 0000007f1c423770
2020-05-06 17:23:13.631 17343-17343/? A/DEBUG: sp 0000007e7ff07050 pc 0000007f1c423798 pstate 0000000060000000
2020-05-06 17:23:13.639 17343-17343/? A/DEBUG: backtrace:
2020-05-06 17:23:13.639 17343-17343/? A/DEBUG: #00 pc 000000000001e798 /system/lib64/libc.so (abort+120)
2020-05-06 17:23:13.639 17343-17343/? A/DEBUG: #1 pc 0000000000054674 /data/app/com.seetatech.fas-Ct8ZuOkPar8bufUuz2hfag==/lib/arm64/libSeetaMaskDetector200_java.so (__gnu_cxx::__verbose_terminate_handler()+348)
2020-05-06 17:23:13.639 17343-17343/? A/DEBUG: #2 pc 0000000000012894 /data/app/com.seetatech.fas-Ct8ZuOkPar8bufUuz2hfag==/lib/arm64/libSeetaMaskDetector200_java.so (__cxxabiv1::__terminate(void (*)())+8)
2020-05-06 17:23:13.639 17343-17343/? A/DEBUG: #3 pc 0000000000012900 /data/app/com.seetatech.fas-Ct8ZuOkPar8bufUuz2hfag==/lib/arm64/libSeetaMaskDetector200_java.so (_ZSt9terminatev+12)
2020-05-06 17:23:13.639 17343-17343/? A/DEBUG: #4 pc 0000000000012a3c /data/app/com.seetatech.fas-Ct8ZuOkPar8bufUuz2hfag==/lib/arm64/libSeetaMaskDetector200_java.so (__cxa_throw+136)
2020-05-06 17:23:13.639 17343-17343/? A/DEBUG: #5 pc 0000000000054e0c /data/app/com.seetatech.fas-Ct8ZuOkPar8bufUuz2hfag==/lib/arm64/libSeetaMaskDetector200.so
2020-05-06 17:23:13.639 17343-17343/? A/DEBUG: #6 pc 0000000000053e08 /data/app/com.seetatech.fas-Ct8ZuOkPar8bufUuz2hfag==/lib/arm64/libSeetaMaskDetector200.so (seeta::v2::MaskDetector::MaskDetector(seeta::ModelSetting const&)+40)
2020-05-06 17:23:13.639 17343-17343/? A/DEBUG: #7 pc 0000000000010fc8 /data/app/com.seetatech.fas-Ct8ZuOkPar8bufUuz2hfag==/lib/arm64/libSeetaMaskDetector200_java.so (test1+1740)
2020-05-06 17:23:13.639 17343-17343/? A/DEBUG: #8 pc 0000000000011a68 /data/app/com.seetatech.fas-Ct8ZuOkPar8bufUuz2hfag==/lib/arm64/libSeetaMaskDetector200_java.so (Java_com_seeta_sdk_MaskDetector_maskDetector+88)
2020-05-06 17:23:13.639 17343-17343/? A/DEBUG: #9 pc 00000000000804a4 /data/app/com.seetatech.fas-Ct8ZuOkPar8bufUuz2hfag==/oat/arm64/base.odex (offset 0x12000)

QualityOfLBN对象不管是动态创建还是new创建，传过去的ModelSetting参数都会出现异常，取不到model数组的值，不知道是不是只有我碰到了。
把头文件的51行QualityOfLBN(const seeta::ModelSetting &setting = seeta::ModelSetting())改为QualityOfLBN( const SeetaModelSetting &setting )、源文件的728行QualityOfLBN::QualityOfLBN(const seeta::ModelSetting &setting)改为QualityOfLBN::QualityOfLBN( const SeetaModelSetting &setting )重新编译后正常使用。

机型：iPhone12
系统版本：iOS 14.4

传入的图片数据经检查没有问题，线程安全问题也已排除。
问题为随机性发生，偶尔会出现。

崩溃在 FaceDetector.h 中：

基于官方的iOS demo做了一个Demo，能够复现该问题，如果没有崩溃，重复启动几次App就会出现。

链接:百度网盘 密码: 3cjs

it is not possible to download the files on the site
https://pan.baidu.com

the site does not accept registration from other countries. It is not possible to receive the SMS code in international numbers.

Is it possible to make the files available elsewhere ??

who knows mega.nz ??

google driver?

使用AVX指令CPU，会加载tennis_sandy_bridge.dll以支持AVX指令。在Intel Core I5 3320M上会抛出错误。

我发现提取的特征有很多零，是因为提取特征后加了relu层，再做归一化吗？请问训练的时候也是加了relu层吗？

这半年你们经历了什么。。。

模型使用的是mask_detector.csta;..so库使用的是libSeetaMaskDetector200.so,运行时报错找不到construct方法

如题

Threshold 调整到 0.05，FaceSize 调整到 80才勉强检测出来部分证件照。

但是与摄像头检测出的人脸比对，相似度又极低。

寻求帮助：加载模型错误 错误：The program '[2840] AlgTest.exe' has exited with code -1073741701
(0xc000007b).

int device_id = 0;
std::string ModelPath = "./models/";
seeta::ModelSetting FD_model;
FD_model.append(ModelPath + "face_detector.csta");
FD_model.set_device(seeta::ModelSetting::CPU);
FD_model.set_id(device_id);

seeta::ModelSetting PD_model;
PD_model.append(ModelPath + "face_landmarker_pts5.csta");
PD_model.set_device(seeta::ModelSetting::CPU);
PD_model.set_id(device_id);

在这里报错： seeta::FaceDetector FD(FD_model); //人脸检测的初始化

作者，您好。麻烦咨询下，后期是否会开放海思NNIE的接口呢？

A/libc: /usr/local/google/buildbot/src/android/ndk-release-r16/external/libcxx/../../external/libcxxabi/src/abort_message.cpp:73: abort_message: assertion "terminating with uncaught exception of type orz::Exception: open the model file:/sdcard/Android/data/jpgparser/face_detector.csta failed!" failed
A/libc: Fatal signal 6 (SIGABRT), code -6 in tid 1984 (AsyncTask #1)

rk3399 上mali GPU 可以跑不？ 出现这个错误：
Compute device gpu has no memory device registered. Please call RegisterMemoryDevice firstly.

能给个Android端人脸识别使用图片注册的Demo吗？原本以为使用openCV将Bitmap转为Mat来提取人脸特征值来注册的，结果好像不行

libSeetaFaceLandmarker600_java.so
libSeetaFaceRecognizer600_java.so
libSeetaFaceDetector600_java.so
Demo中这三个文件只有V7a版本的，在V8a中会直接找不到，感谢提供。

//眼睛状态检测模型初始化
LOGD("Initialize eye_state_detector model: %s\n", eye_state_model_path.c_str());
seeta::ModelSetting eye_state_detector_model;
eye_state_detector_model.set_id(0);
eye_state_detector_model.set_device(SEETA_DEVICE_GPU);
eye_state_detector_model.append(eye_state_model_path);
m_eye_state_detector = new seeta::EyeStateDetector(eye_state_detector_model);

//推理眼睛状态的核心代码
seeta::EyeStateDetector::EYE_STATE left_eye, right_eye;
m_eye_state_detector->Detect(image, points, left_eye, right_eye);

加载模型后,
seeta::FaceAntiSpoofing (anti_setting);
Microsoft C++ 异常: std::length_error

首先非常感谢开放这么好的应用。

我的应用需要准确把握到正脸图片，但是在使用过程中经常会发现把yaw90度的识别为正脸。

使用五点定位时候看到五个关键点分布到了侧脸上，所以通过QulityOfPose检测到是正脸。

使用QulityOfPoseEx有时候也会发现这个问题。

请问下有什么办法可以避免这种情况发生呢？

您好，辛苦，我现在运行centos7 上面出现 【法指令(吐核)】，这个怎样解决呢？windows上的没有问题.楼主的centos7 是哪个版本？

CrashLog:

{
    "exception_reason":"Signal 11 was raised.",
    "exception_name":"UncaughtExceptionHandlerSignalExceptionName",
    "exception_stackInfo":[
        "0   Product                           0x0000000100fc5c44 +[RCCrashManager backtrace] + 60",
        "1   Product                           0x0000000100fc6ecc SignalHandler + 176",
        "2   Product                           0x000000010165d5f8 BLYBSDSignalHandlerForward + 136",
        "3   Product                           0x000000010165d190 BLYBSDSignalHandlerCallback + 152",
        "4   libsystem_platform.dylib          0x00000001a285a9fc <redacted> + 56",
        "5   ???                               0x0000000000000001 0x0 + 1",
        "6   tennis                            0x0000000109033f00 _ZN2ts3cpu13Pooling2DCore9pooling2dERKNS_6TensorENS_13Pooling2DTypeERKNS_8Aspect2DIiEENS_13Padding2DTypeERKNS_6Form2DIiEESE_NS_12Conv2DFormatERS2_ + 3932",
        "7   tennis                            0x0000000108e450b4 _ZN2ts4base9Pooling2D3runERNS_5StackE + 428",
        "8   tennis                            0x0000000108f45648 _ZN2ts11RunOperatorENSt3__110shared_ptrINS_8OperatorEEERNS_5StackEi + 644",
        "9   tennis                            0x0000000109036e98 _ZN2ts3cpu11Pooling2DV23runERNS_5StackE + 1804",
        "10  tennis                            0x0000000108f3bd84 _ZN2ts19OperatorInstruction3runERNS_9WorkbenchE + 704",
        "11  tennis                            0x0000000108f6078c _ZN2ts9Workbench13launch_onlineENSt3__110shared_ptrINS_7ProgramEEEi + 4180",
        "12  tennis                            0x0000000108f5b228 _ZN2ts9Workbench14launch_offlineENSt3__110shared_ptrINS_7ProgramEEERKNS1_6vectorINS_6TensorENS1_9allocatorIS6_EEEE + 300",
        "13  tennis                            0x0000000108f5af04 _ZN2ts9Workbench3runEv + 520",
        "14  tennis                            0x0000000108dff6f0 ts_Workbench_run + 52",
        "15  SeetaFaceDetector600              0x0000000108dac828 _ZNK5seeta2v612FaceDetector9Implement6detec tERK14SeetaImageData + 640",
        "16  SeetaFaceDetector600              0x0000000108da1f5c seeta_v6_FaceDetector_detect + 112",
        "17  Product                           0x0000000101002894 _ZNK5seeta2v612FaceDetector9detect_v2ERK14SeetaImageData + 44",
        "18  Product                           0x0000000101002680 _Z11extract_imgPN5seeta2v612FaceDetectorEPNS0_14FaceRecognizerEPNS0_14FaceLandmarkerENSt3__112basic_stringIcNS7_11char_traitsIcEENS7_9allocatorIcEEEE + 116",
        "19  Product                           0x00000001010043d4 -[SeetaFaceUtil returnFaceFeatureWithIamge:completion:] + 1292"
    ]
}

是否需要建立CMakeLists.txt, 添加库文件的内容怎么写?

如何使用多线程？

terminate called after throwing an instance of 'ts::api::Exception' what(): Compute device gpu has no memory device registered. Please call RegisterMemoryDevice firstly. Aborted (core dumped)

please enlighten me how to fix this problem.

1

error log

mark5  (seetaImageData, face, points)

win10
seetaface6 
cpu: i7-8700k

where to query the error code?

addRule部分

   seeta::QualityAssessor qa;
   qa.add_rule(seeta::INTEGRITY);
   qa.add_rule(seeta::RESOLUTION);
   qa.add_rule(seeta::BRIGHTNESS);
   qa.add_rule(seeta::CLARITY);
   qa.add_rule(seeta::POSE);

disable部分：

   qa.disable(seeta::RESOLUTION);
   qa.disable(seeta::BRIGHTNESS);
   qa.disable(seeta::CLARITY);
   qa.disable(seeta::POSE);

query部分：

   qa.feed(image, face, points, 5);
   auto result = qa.query(seeta::INTEGRITY);

结果抛异常：ERROR: you must add attr before query it.
我把disable改成remove_rule就可以了，请问下disable和remove_rule的区别在哪里

ubuntu18.04编译了源码，然后使用动态库去加载模型，编译通过，但是运行是总是报以下错误：
./demo_face: error while loading shared libraries: libSeetaAuthorize.so: cannot open shared object file: No such file or directory

您好
例如 眼镜检测
seeta::ModelSetting EBD_model("model/eye_state.csta", device, id);
seeta::EyeStateDetector EBD( EBD_model);
因为其他的识别器如五点检测是没有问题的，所以很可能不是个人操作的问题。
具体的错误信息是EBD为null，应该是没加载上模型
期待您看到这条错误信息并给予我回复

0x00007FF941C11420 (vcruntime140.dll)处(位于 HK_Face.exe 中)引发的异常: 0xC0000005: 读取位置 0xFFFFFFFFFFFFFFFF 时发生访问冲突。

使用如下方式，无法载入活体检测模型
seeta::FaceAntiSpoofing *new_fas() { seeta::ModelSetting setting; setting.set_device(seeta::ModelSetting::CPU); setting.append("models/fas_first.csta"); return new seeta::FaceAntiSpoofing(setting); }
显示Segmentation fault
运行环境为ubuntu 18.04，编译链接已链接libSeetaFaceAntiSpoofingX600.so，我之前也已经成功运行人脸检测接口。

其他的评估模块结果正常，只有亮度评估对每个人脸都一样，请问这正常吗？

        auto seetaFaces = ft->FD->detect(simg);
	int featureSize = ft->FR->GetExtractFeatureSize();
	shared_ptr<float> feature(new float[featureSize]);
	SeetaPointF points[5];
	seeta::QualityResult qResult;
	vector<Face> faces;
	for (int i = 0; i < seetaFaces.size; i++)
	{
		ft->FL->mark(simg, seetaFaces.data[i].pos, points);
		ft->FR->Extract(simg, points, feature.get());
		int levels[5];
		float scores[5];
		qResult = ft->QB->check(simg, seetaFaces.data[i].pos, points, 5);
		levels[0] = qResult.level;
		scores[0] = qResult.score;
		qResult = ft->QC->check(simg, seetaFaces.data[i].pos, points, 5);
		levels[1] = qResult.level;
		scores[1] = qResult.score;
		qResult = ft->QI->check(simg, seetaFaces.data[i].pos, points, 5);
		levels[2] = qResult.level;
		scores[2] = qResult.score;
		qResult = ft->QP->check(simg, seetaFaces.data[i].pos, points, 5);
		levels[3] = qResult.level;
		scores[3] = qResult.score;
		qResult = ft->QR->check(simg, seetaFaces.data[i].pos, points, 5);
		levels[4] = qResult.level;
		scores[4] = qResult.score;
		cout << seetaFaces.data[i].pos.x << "," << seetaFaces.data[i].pos.y << "---";
		for (float score : scores)
		{
			cout << score << ", ";
		}
		cout << endl;
		string featureString = "";
		for (int j = 0; j < featureSize; j++)
		{
			featureString += to_string(feature.get()[j]) + ",";
		}
		faces.push_back(Face(imgID, 0, seetaFaces.data[i].pos.x,
							 seetaFaces.data[i].pos.y,
							 seetaFaces.data[i].pos.width,
							 seetaFaces.data[i].pos.height,
							 featureString, levels, scores));
	}

output：

3012,1184---0.0123457, 0.221843, 1, 2.63848, 140, 
2142,1693---0.0123457, 0.366132, 1, 2.74204, 147, 
3351,1627---0.0123457, 0.27993, 1, 2.71851, 150, 
2372,1250---0.0123457, 0.318576, 1, 2.63179, 145, 
2788,1697---0.0123457, 0.396744, 1, 2.90948, 165,

ft->QB是QualityOfBrightness的指针

Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Core was generated by `./example_fas'.
Program terminated with signal SIGILL, Illegal instruction.
#0 0x00007f759fcb6d66 in _GLOBAL__sub_I_conv2d_padding.cpp () from /usr/local/lib/seetaface6/lib64/libtennis.so

iOS端在人脸识别的过程中，发现SeetaFaceDetector600.framework SDK中的FaceDetector.h头文件中的第58行中的 return seeta_v6_FaceDetector_detect(m_impl.get(), &image); 这个函数中包内存地址错误，经测试在这个方法调用之前传入的两个参数均无任何问题。（注意：在iPhone7机型，iOS14.3系统下，问题常现，低于iOS14系统下，偶现次数较少）

作者你好，seetaface6各平台开发包是否可直接免费商用？后续是否考虑开源？

就是我拿自己证件照拍的照片作为图例，给出的 iOS Example 识别出真脸

从摄像头获取图像数据，nv21转bgr，然后brg矩阵传入imageData，imageData传入Detect方法，人脸检测完成；但是当读取单张图片，bitmap转bgr，bgr传入imageData，再传入Detect方法，就检测不到人脸，bgr矩阵转bitmap，然后保存，图片是正常的。但是就是检测不到人脸，请问是为什么，谢谢

hi，您好，
我发现人脸对齐与检测要比原来的seetaface2要慢上不少，不知道是否是我使用的方式不对，原先我的i7大概是10几毫秒，现在要30多毫秒

执行V_SetModelPath报0xc000001d。似乎是指令集不支持。

服务器双GPU：RTX2070；
运行example_FR提示这个错误，请问原因出在哪里

nano下可以直接使用ubuntud的so吗?

既然预留了SEETA_DEVICE_GPU的端口，为什么不提供支持呢？

在Android端调用人脸检测模型时，不管有无人脸，最后返回的结果始终为空，请问可能是哪里出现问题？谢谢

采用的是AVFoundation，想做一个摄像头人像识别，detect的时候报了这个错，请问很可能是什么错？
[ERROR]: [src/backend/base/element_wise_reduce.cpp:53]: [sub:_3] Can not reduce shape: [1, 640, 852, 4] vs. [1, 1, 1, 3]

Telegram group SeetaFace
https://t.me/joinchat/2UXJVSKyn7ZmNTAx