├── LICENSE
├── README.md
└── resource
├── research_fr_fusion.jpg
├── research_fr_overview.jpg
└── research_fr_result_lfw.png
/LICENSE:
--------------------------------------------------------------------------------
1 | GNU LESSER GENERAL PUBLIC LICENSE
2 | Version 3, 29 June 2007
3 |
4 | Copyright (C) 2007 Free Software Foundation, Inc.
5 | Everyone is permitted to copy and distribute verbatim copies
6 | of this license document, but changing it is not allowed.
7 |
8 |
9 | This version of the GNU Lesser General Public License incorporates
10 | the terms and conditions of version 3 of the GNU General Public
11 | License, supplemented by the additional permissions listed below.
12 |
13 | 0. Additional Definitions.
14 |
15 | As used herein, "this License" refers to version 3 of the GNU Lesser
16 | General Public License, and the "GNU GPL" refers to version 3 of the GNU
17 | General Public License.
18 |
19 | "The Library" refers to a covered work governed by this License,
20 | other than an Application or a Combined Work as defined below.
21 |
22 | An "Application" is any work that makes use of an interface provided
23 | by the Library, but which is not otherwise based on the Library.
24 | Defining a subclass of a class defined by the Library is deemed a mode
25 | of using an interface provided by the Library.
26 |
27 | A "Combined Work" is a work produced by combining or linking an
28 | Application with the Library. The particular version of the Library
29 | with which the Combined Work was made is also called the "Linked
30 | Version".
31 |
32 | The "Minimal Corresponding Source" for a Combined Work means the
33 | Corresponding Source for the Combined Work, excluding any source code
34 | for portions of the Combined Work that, considered in isolation, are
35 | based on the Application, and not on the Linked Version.
36 |
37 | The "Corresponding Application Code" for a Combined Work means the
38 | object code and/or source code for the Application, including any data
39 | and utility programs needed for reproducing the Combined Work from the
40 | Application, but excluding the System Libraries of the Combined Work.
41 |
42 | 1. Exception to Section 3 of the GNU GPL.
43 |
44 | You may convey a covered work under sections 3 and 4 of this License
45 | without being bound by section 3 of the GNU GPL.
46 |
47 | 2. Conveying Modified Versions.
48 |
49 | If you modify a copy of the Library, and, in your modifications, a
50 | facility refers to a function or data to be supplied by an Application
51 | that uses the facility (other than as an argument passed when the
52 | facility is invoked), then you may convey a copy of the modified
53 | version:
54 |
55 | a) under this License, provided that you make a good faith effort to
56 | ensure that, in the event an Application does not supply the
57 | function or data, the facility still operates, and performs
58 | whatever part of its purpose remains meaningful, or
59 |
60 | b) under the GNU GPL, with none of the additional permissions of
61 | this License applicable to that copy.
62 |
63 | 3. Object Code Incorporating Material from Library Header Files.
64 |
65 | The object code form of an Application may incorporate material from
66 | a header file that is part of the Library. You may convey such object
67 | code under terms of your choice, provided that, if the incorporated
68 | material is not limited to numerical parameters, data structure
69 | layouts and accessors, or small macros, inline functions and templates
70 | (ten or fewer lines in length), you do both of the following:
71 |
72 | a) Give prominent notice with each copy of the object code that the
73 | Library is used in it and that the Library and its use are
74 | covered by this License.
75 |
76 | b) Accompany the object code with a copy of the GNU GPL and this license
77 | document.
78 |
79 | 4. Combined Works.
80 |
81 | You may convey a Combined Work under terms of your choice that,
82 | taken together, effectively do not restrict modification of the
83 | portions of the Library contained in the Combined Work and reverse
84 | engineering for debugging such modifications, if you also do each of
85 | the following:
86 |
87 | a) Give prominent notice with each copy of the Combined Work that
88 | the Library is used in it and that the Library and its use are
89 | covered by this License.
90 |
91 | b) Accompany the Combined Work with a copy of the GNU GPL and this license
92 | document.
93 |
94 | c) For a Combined Work that displays copyright notices during
95 | execution, include the copyright notice for the Library among
96 | these notices, as well as a reference directing the user to the
97 | copies of the GNU GPL and this license document.
98 |
99 | d) Do one of the following:
100 |
101 | 0) Convey the Minimal Corresponding Source under the terms of this
102 | License, and the Corresponding Application Code in a form
103 | suitable for, and under terms that permit, the user to
104 | recombine or relink the Application with a modified version of
105 | the Linked Version to produce a modified Combined Work, in the
106 | manner specified by section 6 of the GNU GPL for conveying
107 | Corresponding Source.
108 |
109 | 1) Use a suitable shared library mechanism for linking with the
110 | Library. A suitable mechanism is one that (a) uses at run time
111 | a copy of the Library already present on the user's computer
112 | system, and (b) will operate properly with a modified version
113 | of the Library that is interface-compatible with the Linked
114 | Version.
115 |
116 | e) Provide Installation Information, but only if you would otherwise
117 | be required to provide such information under section 6 of the
118 | GNU GPL, and only to the extent that such information is
119 | necessary to install and execute a modified version of the
120 | Combined Work produced by recombining or relinking the
121 | Application with a modified version of the Linked Version. (If
122 | you use option 4d0, the Installation Information must accompany
123 | the Minimal Corresponding Source and Corresponding Application
124 | Code. If you use option 4d1, you must provide the Installation
125 | Information in the manner specified by section 6 of the GNU GPL
126 | for conveying Corresponding Source.)
127 |
128 | 5. Combined Libraries.
129 |
130 | You may place library facilities that are a work based on the
131 | Library side by side in a single library together with other library
132 | facilities that are not Applications and are not covered by this
133 | License, and convey such a combined library under terms of your
134 | choice, if you do both of the following:
135 |
136 | a) Accompany the combined library with a copy of the same work based
137 | on the Library, uncombined with any other library facilities,
138 | conveyed under the terms of this License.
139 |
140 | b) Give prominent notice with the combined library that part of it
141 | is a work based on the Library, and explaining where to find the
142 | accompanying uncombined form of the same work.
143 |
144 | 6. Revised Versions of the GNU Lesser General Public License.
145 |
146 | The Free Software Foundation may publish revised and/or new versions
147 | of the GNU Lesser General Public License from time to time. Such new
148 | versions will be similar in spirit to the present version, but may
149 | differ in detail to address new problems or concerns.
150 |
151 | Each version is given a distinguishing version number. If the
152 | Library as you received it specifies that a certain numbered version
153 | of the GNU Lesser General Public License "or any later version"
154 | applies to it, you have the option of following the terms and
155 | conditions either of that published version or of any later version
156 | published by the Free Software Foundation. If the Library as you
157 | received it does not specify a version number of the GNU Lesser
158 | General Public License, you may choose any version of the GNU Lesser
159 | General Public License ever published by the Free Software Foundation.
160 |
161 | If the Library as you received it specifies that a proxy can decide
162 | whether future versions of the GNU Lesser General Public License shall
163 | apply, that proxy's public statement of acceptance of any version is
164 | permanent authorization for you to choose that version for the
165 | Library.
166 |
--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
1 | ## PIMNet_FR
2 | Created by [Bong-Nam Kang](https://sites.google.com/view/bnkang/home) and [Daijin Kim](http://imlab.postech.ac.kr/members_d.htm) at [POSTECH IM Lab](http://imlab.postech.ac.kr)
3 |
4 | Project page: [https://sites.google.com/view/bnkang/research/face-recognition-using-deep-neural-networks]
5 |
6 |
7 | ### Overview
8 | We propose a new face verification method that uses multiple deep convolutional neural networks (DCNNs) and a deep ensemble, that extracts two types of low dimensional but discriminative and high-level abstracted features from each DCNN, then combines them as a descriptor for face verification. Our DCNNs are built from stacked multi-scale convolutional layer blocks to present multi-scale abstraction. To train our DCNNs, we use different resolutions of triplets that consist of reference images, positive images, and negative images, and triplet-based loss function that maximize the ratio of distances between negative pairs and positive pairs and minimize the absolute distances between positive face images. A deep ensemble is generated from features extracted by each DCNN, and used as a descriptor to train the joint Bayesian learning and its transfer learning method. On the LFW, although we use only 198,018 images and only four different types of networks, the proposed method with the joint Bayesian learning and its transfer learning method achieved 98.33% accuracy. In addition to further increase the accuracy, we combine the proposed method and high dimensional LBP based joint Bayesian method, and achieved 99.08% accuracy on the LFW. Therefore, the proposed method helps to improve the accuracy of face verification when training data is insufficient to train DCNNs.
9 |
10 |
11 | ### Citation
12 |
13 | If you're using this code in a publication, please cite our papers.
14 | ```
15 | @inproceedings{8014823,
16 | author={B. N. Kang and Y. Kim and D. Kim},
17 | booktitle={2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)},
18 | title={Deep Convolutional Neural Network Using Triplets of Faces, Deep Ensemble, and Score-Level Fusion for Face Recognition},
19 | year={2017},
20 | pages={611-618},
21 | doi={10.1109/CVPRW.2017.89},
22 | month={July}
23 | }
24 | ```
25 |
26 |
27 | ### Overall
28 |
29 | Overall procedure of the proposed method. To train deep neural network, we use triplets of faces. With triplets of faces, we train our DCNN with the proposed loss functions to obtain discriminative features. We also train 4 different DCNNs per different resolutions. After training, we extract features from each DCNN model. For test, given face images, these images are passed to multiple DCNNs and then we extract features from each DCNN models. With these extracted features, we classify whether these two face images are same or not using Joint Bayesian Classifier.
30 |
31 |
32 | Hybridization with high-dimensional LBP based joint Bayesian method in the manner of the score-level fusion.
33 |
34 |
35 |
36 |
37 | ### Performance
38 | Comparison of the number of DCNNs, the number of images, the dimensionality of feature, and the accuracy of the proposed method with the state-of-the-art on the LFW.
39 |
40 |
41 |
42 | ### Related Papers
43 |
44 | 1. Bong-Nam Kang, Yonghyun Kim, Daijin Kim, "Deep Convolutional Neural Network using Triplets of Faces, Deep Ensemble, and Score-level Fusion for Face Recognition," 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2017. [[pdf]](http://openaccess.thecvf.com/content_cvpr_2017_workshops/w6/papers/Kang_Deep_Convolutional_Neural_CVPR_2017_paper.pdf) [[slides]](http://vislab.ucr.edu/Biometrics2017/program_slides/PaperID56_CVPRW_BNKANG_Oral.pdf) [[poster]](http://vislab.ucr.edu/Biometrics2017/program_slides/PaperID56_CVPRW_BNKANG_Poster.pdf)
45 | 2. Junghoon Kim, Sang-Seok Yun, Bong-Nam Kang, Daijin Kim, JongSeok Choi, "Reliable Multi-Person Identification using DCNN-based Face Recognition Algorithm and Scale-Ratio Method," 14th International Conference on Ubiquitous Robots and Ambient Intelligence (URAI 2017), 2017. - Best Application Paper Award
46 | 3. 강봉남, 김대진, "공동 손실 함수, Triplet, 스코어 레벨 융합 방법에 의한 Deep 얼굴 인식 방법," 2017 한국컴퓨터종합학술대회 (KCC 2017), 2017. - 우수논문 수상
47 | 4. Bong-Nam Kang, Daijin Kim, "Deep Convolution Neural Network using Triplet of Faces for Face Recognition in the Wild," The 16th International Conference on Control, Automation and Systems (ICCAS 2016), 2016.
48 | 5. Bong-Nam Kang, Daijin Kim, "Deep Convolution Neural Network with Stacks of Multi-Scale Convolutional Layer Block using Triplet of Faces for Face Recognition in the Wild," The 2016 IEEE International Conference on System, Man, and Cybernetics (SMC 2016), 2016.
49 |
50 |
51 | ### Code
52 |
53 | Download the executalable binary file: [FR_PIMNet_v2.0.zip](http://imlab.postech.ac.kr/software/FR_PIMNET_v2.0.zip)
54 |
55 | #### System Requirements
56 | * This software is tested on Microsoft Windows 7 and 8.1 (64bit).
57 | * At least 6GB gpu memory is required (NVIDIA Titan black, X, and Tesla K20 are used for testing).
58 | #### Dependencies
59 | * NVIDIA CUDA 8.0, cuDNN 5.0
60 | * OpenCV 3.1
61 | * Caffe-windows
62 |
63 |
64 | ### Licence
65 |
66 | This software is for research purpose only.
67 |
68 | Check LICENSE file for details.
69 |
70 |
71 | ### Acknowledgements
72 |
73 | This research was supported by the MSIT (Ministry of Science, ICT), Korea, under the SW Starlab support program (IITP-2017-0-00897) supervised by the IITP (Institute for Information & communications Technology Promotion) and also supported by the MSIT, Korea, under the ICT Consilience Creative program (IITP-2017-R0346-16-1007) supervised by the IITP.
74 |
75 |
--------------------------------------------------------------------------------
/resource/research_fr_fusion.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/POSTECH-IMLAB/PIMNet_FR-1/d7a8e1d6d36b46b01db1b1d47cbc883f7b0e0104/resource/research_fr_fusion.jpg
--------------------------------------------------------------------------------
/resource/research_fr_overview.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/POSTECH-IMLAB/PIMNet_FR-1/d7a8e1d6d36b46b01db1b1d47cbc883f7b0e0104/resource/research_fr_overview.jpg
--------------------------------------------------------------------------------
/resource/research_fr_result_lfw.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/POSTECH-IMLAB/PIMNet_FR-1/d7a8e1d6d36b46b01db1b1d47cbc883f7b0e0104/resource/research_fr_result_lfw.png
--------------------------------------------------------------------------------