Pytorch_Face_Recognition-
Pytorch_Face_Recognition- copied to clipboard
Pytorch implementation of mainstream face recognition algorithms(ArcFace, CosFace).
out_feature输出样本的大小,是否是数据集num的大小?比如MS_Celeb_1M应该是85164吗?
when reproducing the model, I used retina to detect the landmarks and aligned by mtcnn, crop to 112 on the CASIA and LFW, the accuracy on LFW is 98.xx %,...
感谢您出色的代码和模型! 我下载了您的mobilenet和SeResnet50两个预训练模型,并利用自己的数据进行了测试,我将自己的正面证件照混在600个其他人的人脸图片中,并利用一张自己的自拍照作为测试图。 对于601张照片,我先使用RetinaFace模型将所有人脸进行截取与对齐,并压缩至112,112尺寸,该操作同样被用于自拍的测试图。 我利用预训练模型对所有照片进行推理,得到[601,128],[1,128]两个矩阵,并分别利用sklearn提供的cosine_similarity,euclidean_distances函数,求得两个相似度矩阵,无论使用这两种方法的哪一种方法,目前我都无法将我的自拍照与我的证件照正确匹配(相似度最高),请问我的思路与方法是否错误,或者仍有什么地方需要改进? ps:所有照片的分辨率均是在3000*4000以上
could you provide the cleaned aligned 112*112 datasets? like baidu cloud link
Does anyone know how to draw roc?
hello,i wanna to know your lfw.py and lfw_test_pairs.txt are right? i got this error, ValueError: invalid literal for int() with base 10: 'Akhmed_Zakayev/Akhmed_Zakayev_0003.jpg' the lfw_test_pairs.txt 
I used face alignment to original CASIA and LFW to prepare aligned datasets using code from https://github.com/ZhaoJ9014/face.evoLVe.PyTorch (align/face_align.py). However, after training, my accuracy on LFW recorded 96.4833 %. When I...
Hi, I appreciate your hard work, and I want to use your well-organized code for research purposes. But there's a problem with connecting the 'pan-baidu' server when I tried to...