DeepFace
所属分类：图形图像处理
开发工具：Python
文件大小：35113KB
下载次数：86
上传日期：2016-11-25 09:28:06
上 传 者：
齐哥
说明： DeepFace一文依旧是沿着“检测-对齐-人脸表示-分类”这一人脸识别技术路线来的，其贡献在于对人脸对齐和人脸表示环节的改进。1）在人脸对齐环节，引入了3D人脸模型对有姿态的人脸就行分片的仿射对齐。2）在人脸表示环节，利用一个9层的深度卷积在包含4000人、400万张人脸的数据集上学习人脸表示，这个9层的DCNN网络有超过1.2亿个参数。本文的模型在LFW数据集上取得了97.25 的平均精度（逼近了人类97.5 的极限），同时在Youtube数据集上取得了当前最好的结果，比之前的NO.1整整高出了12.7 。
(DeepFace article is still along the detection- alignment- face representation- classification of this face recognition technology line, its contribution to the face alignment and face representation of the improvement. 1) In the face alignment part, the 3D human face model is introduced to affine alignment of slices with face. 2) In the face representation session, using a 9-layer depth convolution on a data set containing 4,000 human faces and 4 million human faces, the 9-layer DCNN network has more than 120 million parameters. The model obtained 97.25 average precision (approximating the 97.5 limit of human) on the LFW dataset, and achieved the best result on the Youtube dataset, 12.7 higher than the previous NO.1)
文件列表:
FaceAlignment (0, 2016-04-14)
FaceAlignment\DataPrecessing (0, 2016-04-14)
FaceAlignment\DataPrecessing\faceProcessing.py (4533, 2016-04-14)
FaceAlignment\DataPrecessing\imagelist2hdf5.py (2785, 2016-04-14)
FaceAlignment\Evaluate (0, 2016-04-14)
FaceAlignment\Evaluate\demo.py (2081, 2016-04-14)
FaceAlignment\Evaluate\evaluate.py (2377, 2016-04-14)
FaceAlignment\Evaluate\recovery_all.py (1501, 2016-04-14)
FaceAlignment\Evaluate\result (0, 2016-04-14)
FaceAlignment\Evaluate\result\1.jpg (1693, 2016-04-14)
FaceAlignment\Evaluate\result\10.jpg (1781, 2016-04-14)
FaceAlignment\Evaluate\result\100.jpg (1692, 2016-04-14)
FaceAlignment\Evaluate\result\101.jpg (1718, 2016-04-14)
FaceAlignment\Evaluate\result\102.jpg (1645, 2016-04-14)
FaceAlignment\Evaluate\result\103.jpg (1655, 2016-04-14)
FaceAlignment\Evaluate\result\104.jpg (1695, 2016-04-14)
FaceAlignment\Evaluate\result\105.jpg (1935, 2016-04-14)
FaceAlignment\Evaluate\result\106.jpg (1761, 2016-04-14)
FaceAlignment\Evaluate\result\107.jpg (1779, 2016-04-14)
FaceAlignment\Evaluate\result\108.jpg (1722, 2016-04-14)
FaceAlignment\Evaluate\result\109.jpg (1549, 2016-04-14)
FaceAlignment\Evaluate\result\11.jpg (1612, 2016-04-14)
FaceAlignment\Evaluate\result\110.jpg (1699, 2016-04-14)
FaceAlignment\Evaluate\result\111.jpg (1702, 2016-04-14)
FaceAlignment\Evaluate\result\112.jpg (1733, 2016-04-14)
FaceAlignment\Evaluate\result\113.jpg (1730, 2016-04-14)
FaceAlignment\Evaluate\result\114.jpg (1733, 2016-04-14)
FaceAlignment\Evaluate\result\115.jpg (1716, 2016-04-14)
FaceAlignment\Evaluate\result\116.jpg (1649, 2016-04-14)
FaceAlignment\Evaluate\result\117.jpg (1710, 2016-04-14)
FaceAlignment\Evaluate\result\118.jpg (1834, 2016-04-14)
FaceAlignment\Evaluate\result\119.jpg (1680, 2016-04-14)
FaceAlignment\Evaluate\result\12.jpg (1597, 2016-04-14)
FaceAlignment\Evaluate\result\120.jpg (1597, 2016-04-14)
FaceAlignment\Evaluate\result\121.jpg (1740, 2016-04-14)
FaceAlignment\Evaluate\result\122.jpg (1720, 2016-04-14)
FaceAlignment\Evaluate\result\123.jpg (1672, 2016-04-14)
FaceAlignment\Evaluate\result\124.jpg (1642, 2016-04-14)
FaceAlignment\Evaluate\result\125.jpg (1604, 2016-04-14)
... ...
# DeepFace
a deep face analysis implement, mainly based on -[Caffe](https://github.com/BVLC/caffe). At this time, face analysis tasks like detection, alignment and recognition have been done. -[中文Readme](https://github.com/RiweiChen/DeepFace/blob/master/README_ch.md)
Each task is divide by different folder.
##Face detection
###baseline
Face detection using sliding windows style, it first train a face/noface two class classification network, and then tranform to a full convolution network, detection is based on the heatmap when input a large full size image.
face detection result example:
##Face key points detection
###try1_1
face 5 key points detection using DeepID architecture.
face 5 key points detection result example:
##Face Verification
###deepid
Face verification based on DeepID network architecture.
face verification ROC result:
##Face Dataset
We collect the face datasets usually used by recently years' paper, and divide by different task. Also we simply describe each of the dataset.(In Chinese)
For more implement details, please reference my blog
- [1***3的专栏](http://blog.csdn.net/chenriwei2)
It is welcome for everyone to make suggestions and improvement.
近期下载者：
相关文件：
收藏者：