DeepID-implementation

DeepID-implementation is an implementation of paper "Deep Learning Face Representation from Predicting 10,000 Classes", which proposes to learn a set of compact, 160-dims high level feature representations through deep learning, referred to as Deep hidden IDentity features (DeepID), for face verification.

More details in DeepID Notebook.

Dataset

Dataset	People	Image	Size
CASIA-WebFace-custom	10,575	392,304	62x62
LFW	6,000 pairs		62x62

Train, Test Model

Model	People	Training images	Validation images	Train	Test
62x62_3_e2_0264	264	9,554	939	train	test
62x62_3_e2_0528	528	16,076	1,544	train	test
62x62_3_e2_1057	1,057	37,977	3,686	train	test
62x62_3_e2_2115	2,115	72,126	6,978	train	test
62x62_3_e2_4230	4,230	143,939	13,977	train	test
62x62_3_e2_8460	8,460	284,466	27,498	train	test
62x62_3_e1_conventional	8,460	284,466	27,498	train	test

Experiments

• The classification ability of Multi-scale ConvNets
  • [paper] | [model_62x62_3]
• The effectiveness of the learned hidden representations for face verification
  • [paper] | [model_62x62_3]
• The learned features extract identity information
  • [paper] | [model_62x62_3]
• Various face patches combination contributes to the performance
  • [paper]
• Method comparison
  • [paper] | [model_62x62_3]

Test accuracy on LFW

with Joint Bayesian, Cosine Similarity, Euclidean Distance method.

Model	Joint Bayesian	Cosine Similarity	Euclidean Distance
62x62_3_e2_0264	0.61	0.735	0.6745
62x62_3_e2_0528	0.60	0.749333	0.701
62x62_3_e2_1057	0.61	0.758333	0.709167
62x62_3_e2_2115	0.65	0.778167	0.7365
62x62_3_e2_4230	0.66	0.793167	0.756833
62x62_3_e2_8460	0.66	0.7985	0.757
62x62_3_e1_conventional	0.64	0.797667	0.763167

Reference

[1]. Yi Sun, Xiaogang Wang, Xiaoao Tang, Deep Learning Face Representation from Predicting 10,000 Classes, 2014-06-23

[2]. RiweiChen, Caffe实践-基于Caffe的人脸识别实现, 2015-11-01

[3]. RiweiChen, 深度学习论文笔记-Deep Learning Face Representation from Predicting 10,000 Classes, 2014-06-16

[4]. 張雨石, DeepID人脸识别算法之三代, 2014-12-23

[5]. RiweiChen, RiweiChen/DeepFace, 2016-04-14

[6]. Feng Wang, happynear/FaceVerification, 2016-04-25

[7]. Alfred Xiang Wu, AlfredXiangWu/face_verification_experiment, 2015-12-14