# Dataset之LFW：LFW人脸数据库的简介、安装、使用方法之详细攻略 ###### tags:`學習紀錄` [toc] --- ## Before Meeting :::success - A database of face photographsdesigned for studying the problem of unconstrained face recognition.The data set contains more than 13,000 images of faces collected fromthe web. Each face has been labeled with the name of the personpictured. 1680 of the people pictured have two or more distinct photosin the data set. The only constraint on these faces is that they weredetected by the Viola-Jones face detector. More details can be foundin the technical report below.    There are now four different sets of LFW images including the original and three different types of "aligned" images.The aligned images include "funneled images" (ICCV 2007), LFW-a, which uses an unpublished method of alignment,and "deep funneled" images (NIPS 2012). Among these, LFW-a and the deep funneled images produce superior results for most face verification  algorithms over the original images and over the funneled images (ICCV 2007). ::: [refer](https://blog.csdn.net/qq_41185868/article/details/82915063) [refer](https://blog.csdn.net/jobbofhe/article/details/79416661) [refer]() --- ## Recent Paper --- ### LFW :::success #### Abstracion ::: :::info #### Detail - LFW（人脸比对数据集） - ![](https://i.imgur.com/pzuibth.png) - 测试过程概述 - 通过dlib进行人脸识别网络训练后，得到dlib_face_recognition_resnet_model_v1.dat。通常大家在LFW人脸数据集上对该模型进行精度验证。以下梳理验证过程： - 在原始LFW数据集中，截取人脸图像并保存。（例如：可以使用开源人脸检测对齐seetaface将人脸crop出来，并 保存，建议以原图像名称加一个后缀命名人脸图像） - 通过python，matlab，或者C++，构建训练时的网络结构并加载dlib_face_recognition_resnet_model_v1.dat。 - 将截取的人脸送入网络，每个人脸都可以得到网络前向运算的最终结果，一般为一个N维向量，并保存，建议以原图像名称加一个后缀命名。 - LFW提供了6000对人脸验证txt文件，lfw_pairs.txt,其中第1个300人是同一个人的两幅人脸图像；第2个300人是两个不同人的人脸图像。按照该list，在（3）保存的数据中，找到对比人脸对应的N维特征向量。 - 通过cosine距离/欧式距离计算两张人脸的相似度。同脸和异脸分别保存到各自对应的得分向量中。 - 同脸得分向量按照从小到大排序，异脸向量按照从大到小排序。 - FAR（错误接受率）从0~1，按照万分之一的单位，利用排序后的向量，求FRR（错误拒绝率）或者TPR（ture positive ratio）。 - 根据7可绘制ROC曲线。 ::: :::warning #### Conclusion ::: [refer]() --- :::success #### Abstracion ::: :::info #### Detail ::: :::warning #### Conclusion ::: [refer]() ---