seetaface / seetafaceengine2 Goto Github PK

@seetaface Hi, Thanks for the libraries. I was testing the detection lib with few images. One of the issues i see it doesn't detect small faces. Is there any way to pass any argument to detect the small faces. Thanks

here is the image it detect only few faces.
https://ibb.co/h729RSs

I compare it with RetinaFace detection
https://ibb.co/p33JF0B

检测彩色图的时候能够准确的识别出来，但同样的图片变成黑白图的时候就检测不出来了

Hi， I tested several times, and found that face registration only supports ten faces, and then it will not recognize. Why?

VS2013 + opencv249 + SeetaFaceEngine2 编译出现Error C2872:: "cv" 不明确的符号，检查发现是seeta/Struct_cv.h这个文件重定义了namespace cv导致，请问这问题怎么解决？

希望能提供Android、Windows、Linux的编译说明。
另本人使用的是Mac OS X，如果Linux能编译的话，理论上Mac OS X应该也是能编译的。

在使用seeta::FaceRecognizer2 FR("bindata/SeetaFaceRecognizer2.0.ats");时。
编译后会提示，

I would like to edit some lines of code. Is it possible to get the cpp src file?
Thank you

Hi all, thank again for your great contribution, could you kindly share the training scripts? thank you

com.seeta.sdk.PointDetector2运行错误
Error：No implementation found for void com.seeta.sdk.PointDetector2.construct(java.lang.String) (tried Java_com_seeta_sdk_PointDetector2_construct and Java_com_seeta_sdk_PointDetector2_construct__Ljava_lang_String_2)

com.seeta.sdk.FaceDetector2运行正常

您好，请问Android下初始化需要的faceDetectorModel，pointDetectorModel，faceRecognizerModel在哪里下载

你好，我看头文件里似乎只支持vs2013和vs2015？

Great work team!
Can you please provide models for other architectures of android systems like "armeabi", "armeabi-v8"?

As far as I know, the number of photos registered is limited to 500. So I want to know, if the number of photos that I would compare to is more than 500, for example, 8000, then, is there a way to process 16 groups of 500 photos simultaneously? At first, I wanted to use the multi threads to process, however, I found that the class FaceRecognizer2 seems like only to be instantiated once a time.
Is it right? Hope to get your answers.

Is there something like "ExtractFeatureWithCrop()" in version 1.0?

Hi,

Does the pretrained weight can be used for commercial use? or just the header files are under BSD license?

1.文档中指出安卓端人脸检测HUAWEI meta8上可达到37FPS，实测高通625芯片，仅8FPS；
2.检测FDDB曲线不错，实测中对眼镜非常敏感，误检率很高；
3.另刚更新的安卓端人脸特征点so库初始化错误，无法运行，第一版正常。
4.minSdkVersion支持的最低版本？
以上是个人实测发现的小问题，不知道是否仅个人情况。

Hi guys,

I create this repo for the one who is looking for python version. Feel free to open any issues:)

Hi, it's really great job. Runs four times faster than Seeta 1 and much more accurate!
Do you plan to share the sources? (Or maybe sell them :) )?

目前使用的opencv版本是4.1.0，引入项目后提示缺少成员imread

之前曾经提过会开放无限人脸数据库版本的sdk的申请渠道，请问要如何申请？以及去哪申请？抑或是还没有开放申请吗？

Hi, thanks for the great job. It works pretty cool, fast and rather accurate.
If no sources provided, could you provide builds for Ubuntu 14?

Thanks!

为什么同一张照片的相似度是0.999999，而不是1呢？

I've tested SeetaFaceEngine2 on ubuntu 16.04 64bit. The result is pretty good.
Wondering if the sdk supports 32 bit system, which is the most case for embedded systems?
If not, can you publish 32 bit sdk?

你好，感谢你们的开源，我在运行seetaface2时发现人脸库最多只能添加10个，请问可以增加人脸库的添加数量吗

HI，
我用JNA调用X86版本，发现调用SeetaFacePointDetector200.dll失败 具体提示是
找不到函数。 发现 就这个构造函数是输出的时候没有重命名接口函数过的。
是不是导出接口时候 需要对这个函数重命名下。

java.lang.UnsatisfiedLinkError: dlopen failed: cannot locate symbol "atof" referenced by "libFaceDetectorJni.so"...

Hi
I am downloading this model file from Baidu. It shows me the data size about 124MB but I can download successfully only 32MB.

1. Is there any mistake here?
2. because the downloaded file is ".exe", I do not know how to convert exe into ".ats"?

Could you please help me? Thanks

cv::Mat mat2 = cv::imread(imageB_Path.c_str());
seeta::cv::ImageData image2 = mat2;
int probe_face_num = 0 ;
SeetaRect *face2 = FD.Detect(image2, &probe_face_num);
SeetaPointF points2[5];
PD.DetectEx(image2, *face2, points2);//这一句程序退出
图片如下：

我整好后发现这俩好像并不是一种东西。。。

Hi,
the link for downloading MD5 model is broken, could you please help update again?

测试了不同质量的图片,质量差的会直接段错误,这种程度的错误恐怕需要一个说明文档

在Android端测试发现，人脸检测与标记比SeetaFaceEngine快，但是

float similarity = faceRecognizer.Compare(image1, landmarks1, image2, landmarks2);

比SeetaFaceEngine要慢很多，而

float[] similarity = new similarity[1];//save the most similar face similarity value
int targetIndex = faceRecognizer.Recognize(image, landmarks, similarity);//targetIndex is the index of the most similar face in database

和SeetaFaceEngine差不多。但是这个database有上限。

请问能否提高faceRecognizer.Compare(image1, landmarks1, image2, landmarks2);速度

hello. Thank you for sharing this awesome project. Is there a way to share also the build for msvc 2015, or the source code? Thank you.

如果是一直检测人脸，那么就要一直new bitmap，这个在Android中是非常消耗资源的，能否直接检测byte[] 数据流，而不是bitmap

Hi, Does seetafaceengine2 support running on armv7 platform such as rk3288? I found the so library are build on x86 but not found the library build on armv7.

Is there anybody trying to use the code in native level for Android Developing? I need some guidances

Are there any plans to support the SDK for ARM Architecture?

发现这个，是不是可以设置GPU上跑该程序
enum SeetaDevice
{
SEETA_DEVICE_AUTO,
SEETA_DEVICE_CPU,
SEETA_DEVICE_GPU,
}

Can you give me another link downloader except for Baidu?

除了人脸库限制10人，程序还有那些坑希望厂家说明！

it failed to create multiple instance of model instances of PointDetector. program crashed after second instance

您好，请教Android下使用库的2个问题。 1、face registration后，生成的数据库是在Android下哪个目录
2、为什么每次APP退出后重新开始，之前注册在数据库的信息就不见了，再注册时产生的索引又是从0开始

首先感谢分享。
简单跑了下SeetaFaceEngine： https://github.com/seetaface/SeetaFaceEngine
效果相当差。用Lfw中的一个人的两张图，相似度能达到0.89,但从网上随便下载同一个人两张图（奥巴马，普金），相似度只有零点二几。用自己的两张照片也只有0.4+.

有人跑过这个库识别效果怎样？

Hi, on face alignment, the model you used predicts 68 landmarks, but this code can only predict 5, so how can I predict 68 landmarks?