3000fps&SDM方法资源摘录

去年在美图公司实习的时候,研究了一段时间的SDM方法,写了一系列的博客,见《Supervised Descent Method and its Applications to Face Alignment》等,今年来到了杭州,进入了图片社交领域的佼佼者in实习,主要还是做人脸对齐。最近一段时间,一直在研究《Face Alignment at 3000 FPS via Regressing Local Binary Features》这篇文章,也为此搜集了很多资料,特整理在此,以供参考。

paper:

3000fps论文链接:

github code linking:

3000fps:

SDM:

other linking

开源库链接及评价

  1. dlib :https://github.com/davisking/dlib/tree/v18.18
    评价:速度快,可商用,有些时候不太准确
  2. CLM-framework: https://github.com/TadasBaltrusaitis/CLM-framework,
    新版已经改为OpenFace,见:https://github.com/TadasBaltrusaitis/OpenFace
    评价:很准确,不可商用
  3. Face Detection, Pose Estimation and Landmark Localization in the Wild :http://www.ics.uci.edu/~xzhu/face/
    评价:Very slow (~10 seconds an image after hyper threading on a 8-core CPU), but very accurate when it comes to high pose variations
  4. SDM patrikhuber/superviseddescent:https://github.com/patrikhuber/superviseddescent
    评价:Nicely written C++ code, though not very robust
  5. Robust face landmark estimation under occlusion:http://www.vision.caltech.edu/xpburgos/ICCV13/
    评价:Specially designed for handling occlusions(遮挡区域), but slow on account being written in MATLAB.
  6. 应用了CLM的项目:https://www.technologyreview.com/s/541866/this-car-knows-your-next-misstep-before-you-make-it/
    评价:I actually explored a large number of open-source facial landmark detectors for a project, and found the CLM framework to outperform everything else (in terms of both speed and accuracy). We eventually used it in our project: www.technologyreview.com/news/…
  7. clandmark:https://github.com/uricamic/clandmark
  8. kylemcdonald/FaceTracker:https://github.com/kylemcdonald/FaceTracker
  9.  Android app for facial landmark tracking
    评价:安装该app需要OpenCV Manager,我已提供链接.。
    至于效果,不是很好,需要优化。不过上面提供了编写jni的代码,对于编写C++的API应该会有帮助。
  10. kylemcdonald/FaceTracker:https://github.com/kylemcdonald/FaceTracker
  11. ofxFaceTracker:https://github.com/kylemcdonald/ofxFaceTracker

参考文献

Facial Landmark Detection

Enjoy it ? Donate me ! 欣赏此文?求鼓励,求支持!