Deep learning models are shown to be vulnerable to adversarial examples. Though adversarial train... more Deep learning models are shown to be vulnerable to adversarial examples. Though adversarial training can enhance model robustness, typical approaches are computationally expensive. Recent works proposed to transfer the robustness to adversarial attacks across different tasks or models with soft labels. Compared to soft labels, feature contains rich semantic information and holds the potential to be applied to different downstream tasks. In this paper, we propose a novel approach called Guided Adversarial Contrastive Distillation (GACD), to effectively transfer adversarial robustness from teacher to student with features. We first formulate this objective as contrastive learning and connect it with mutual information. With a well-trained teacher model as an anchor, students are expected to extract features similar to the teacher. Then considering the potential errors made by teachers, we propose sample reweighted estimation to eliminate the negative effects from teachers. With GACD, the student not only learns to extract robust features, but also captures structural knowledge from the teacher. By extensive experiments evaluating over popular datasets such as CIFAR-10, CIFAR-100 and STL-10, we demonstrate that our approach can effectively transfer robustness across different models and even different tasks, and achieve comparable or better results than existing methods. Besides, we provide a detailed analysis of various methods, showing that students produced by our approach capture more structural knowledge from teachers and learn more robust features under adversarial attacks.
Large-scale, highly integrated and low-power-consuming hardware is becoming progressively more im... more Large-scale, highly integrated and low-power-consuming hardware is becoming progressively more important for realizing optical neural networks (ONNs) capable of advanced optical computing. Traditional experimental implementations need N2 units such as Mach-Zehnder interferometers (MZIs) for an input dimension N to realize typical computing operations (convolutions and matrix multiplication), resulting in limited scalability and consuming excessive power. Here, we propose the integrated diffractive optical network for implementing parallel Fourier transforms, convolution operations and application-specific optical computing using two ultracompact diffractive cells (Fourier transform operation) and only N MZIs. The footprint and energy consumption scales linearly with the input data dimension, instead of the quadratic scaling in the traditional ONN framework. A ~10-fold reduction in both footprint and energy consumption, as well as equal high accuracy with previous MZI-based ONNs was ...
2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021
This paper studies the problem of panoramic image reflection removal, aiming at reliving the cont... more This paper studies the problem of panoramic image reflection removal, aiming at reliving the content ambiguity between reflection and transmission scenes. Although a partial view of the reflection scene is included in the panoramic image, it cannot be utilized directly due to its misalignment with the reflection-contaminated image. We propose a two-step approach to solve this problem, by first accomplishing geometric and photometric alignment for the reflection scene via a coarse-to-fine strategy, and then restoring the transmission scene via a recovery network. The proposed method is trained with a synthetic dataset and verified quantitatively with a real panoramic image dataset. The effectiveness of the proposed method is validated by the significant performance advantage over single image-based reflection removal methods and generalization capacity to limited-FoV scenarios captured by conventional camera or mobile phone users
Complex-valued neural networks have many advantages over their real-valued counterparts. Conventi... more Complex-valued neural networks have many advantages over their real-valued counterparts. Conventional digital electronic computing platforms are incapable of executing truly complex-valued representations and operations. In contrast, optical computing platforms that encode information in both phase and magnitude can execute complex arithmetic by optical interference, offering significantly enhanced computational speed and energy efficiency. However, to date, most demonstrations of optical neural networks still only utilize conventional real-valued frameworks that are designed for digital computers, forfeiting many of the advantages of optical computing such as efficient complex-valued operations. In this article, we highlight an optical neural chip (ONC) that implements truly complex-valued neural networks. We benchmark the performance of our complex-valued ONC in four settings: simple Boolean tasks, species classification of an Iris dataset, classifying nonlinear datasets (Circle a...
Gait recognition is an emerging biometric technology that identifies people through the analysis ... more Gait recognition is an emerging biometric technology that identifies people through the analysis of the way they walk. The challenge of model-free based gait recognition is to cope with various intra-class variations such as clothing variations, carrying conditions and angle variations that adversely affect the recognition performance. This paper proposes a method to select the most discriminative human body part based on group Lasso of motion to reduce the intra-class variation so as to improve the recognition performance. The proposed method is evaluated using CASIA Gait Dataset B. Experimental results demonstrate that the proposed technique gives promising results.
Semantic image segmentation aims to classify every pixel of a scene image to one of many classes.... more Semantic image segmentation aims to classify every pixel of a scene image to one of many classes. It implicitly involves object recognition, localization, and boundary delineation. In this paper, we propose a segmentation network called CGBNet to enhance the segmentation performance by context encoding and multi-path decoding. We first propose a context encoding module that generates context-contrasted local feature to make use of the informative context and the discriminative local information. This context encoding module greatly improves the segmentation performance, especially for inconspicuous objects. Furthermore, we propose a scale-selection scheme to selectively fuse the segmentation results from different-scales of features at every spatial position. It adaptively selects appropriate score maps from rich scales of features. To improve the segmentation performance results at boundary, we further propose a boundary delineation module that encourages the location-specific verylow-level features near the boundaries to take part in the final prediction and suppresses them far from the boundaries. The proposed segmentation network achieves very competitive performance in terms of all three different evaluation metrics consistently on the six popular scene segmentation datasets, Pascal Context, SUN-RGBD, Sift Flow, COCO Stuff, ADE20K, and Cityscapes.
• Identification requires comparison of a person’s fingerprint with all fingerprints in a fingerp... more • Identification requires comparison of a person’s fingerprint with all fingerprints in a fingerprint database. Most of existing fingerprint datasets are very large (>1 million fingerprints). • In some cases other attributes such as race, gender, age, soft biometrics are introduces in the database, which allows partitioning of the dataset into smaller subsets. However, in most cases the data are fingerprints only. • Comparison of an on-line acquired fingerprint with all fingerprints in the database is computationally expensive (FBI dataset has >200 million fingerprints). • A solution to this problem is to divide the database into a number of bins (based on some predefined automatically extracted general fingerprint features). • Classification is referred to the assigning a class in consistent and reliable way. • Fingerprint matching is based on local features while fingerprint classification is based on global features, such as global ridge patter and singularities.
ABSTRACT Due to high dimensionality of images or generated color features, different color channe... more ABSTRACT Due to high dimensionality of images or generated color features, different color channels are usually processed separately and then concatenated together into a feature vector for classification. This makes channel fusion a crucial step in color face recognition (FR) systems. However, existing methods simply concatenate channel-wise color features without identifying the importance or reliability of features in different color channels. In this paper, we propose a color channel fusion (CCF) approach using jointly dimension reduction algorithms to select more features from reliable and discriminative channels. Experiments using two different dimension reduction approaches, two different types of features on three image datasets show that CCF achieves consistently better performance than color channel concatenation (CCC) method which deals with different color channels equally.
In this paper, we address the false rejection problem due to small solid state sensor area availa... more In this paper, we address the false rejection problem due to small solid state sensor area available for fingerprint image capture. We propose a minutiae data synthesis approach to circumvent this problem. Main advantages of this approach over existing image mosaicing approach include low memory storage requirement and low computational complexity. Moreover, possible overhead on the search engine (for fingerprint matching) due to data redundancy could be reduced. Extensive experiments were conducted to determine the best transformation suitable for minutiae alignment. We demonstrate the idea of synthesis with an example using physical fingerprint images. The proposed synthesis system is also found to improve (lower) the number of false rejects due to the use of different fingerprint regions for matching.
This paper proposes a novel full-reference quality assessment (QA) metric that automatically asse... more This paper proposes a novel full-reference quality assessment (QA) metric that automatically assesses the quality of an image in the discrete orthogonal moments domain. This metric is constructed by representing the spatial information of an image using low order moments. The computation, up to fourth order moments, is performed on each individual ð8 Â 8Þ non-overlapping block for both the test and reference images. Then, the computed moments of both the test and reference images are combined in order to determine the moment correlation index of each block in each order. The number of moment correlation indices used in this study is nine. Next, the mean of each moment correlation index is computed and thereafter the single quality interpretation of the test image with respect to its reference is determined by taking the mean value of the computed means of all the moment correlation indices. The proposed objective metrics based on two discrete orthogonal moments, Tchebichef and Krawtchouk moments, are developed and their performances are evaluated by comparing them with subjective ratings on several publicly available databases. The proposed discrete orthogonal moments based metric performs competitively well with the state-of-the-art models in terms of quality prediction while outperforms them in terms of computational speed.
Discussion papers are research materials circulated by their authors for purposes of information ... more Discussion papers are research materials circulated by their authors for purposes of information and discussion. They have not necessarily undergone formal peer review.
ABSTRACT JIANG X., DIETZENBACHER E. and LOS B. Improved estimation of regional input-output table... more ABSTRACT JIANG X., DIETZENBACHER E. and LOS B. Improved estimation of regional input-output tables using cross-regional methods, Regional Studies. Many regional input-output tables are estimated by means of non-survey methods. Often, information on the margins of the projected table is complemented by full information on intermediate inputs from tables for other regions. This paper compares the performance of four of such 'cross-regional' methods. Two of these were already proposed in the literature, whereas the other two are based on recent advances in regression analysis. The methods are tested not only against each other, but also against traditional methods that do not employ cross-regional information. To this end, twenty-seven regional input-output tables for China in 1997 and 2002 are used.
Beijing Da Xue Xue Bao Yi Xue Ban Journal of Peking University Health Sciences, Jul 1, 2003
To investigate the feasibility and cortex activation of fMRI in aged volunteers during the perfor... more To investigate the feasibility and cortex activation of fMRI in aged volunteers during the performance of calculation tasks. Internal mental calculation tasks including simple tasks and complicated tasks were conducted in 11 normal aged volunteers. The fMRI data was postprocessed using SPM99 to reveal the activated cortex. The simple calculation tasks were fulfilled satisfactorily in all aged subjects. Cortex in the bilateral Superior parietal lobule, inferior parietal lobule and bilateral occipital lobes showed activation. We found a selective enhancement of fMRI signal in Brodmann regions 9, 10 and 46 in the complicated tasks. fMRI tasks are feasible in aged people and more cortex of the left frontal lobe shows activation in complicated calculation tasks.
Http Dx Doi Org 10 1080 09500340 2012 698659, Oct 10, 2012
The characterization and analysis of afterpulsing behavior in InGaAs/InP single photon avalanche ... more The characterization and analysis of afterpulsing behavior in InGaAs/InP single photon avalanche diodes (SPADs) is reported for gating frequencies between 10 and 50 MHz. Gating in this frequency range was accomplished using a matched delay line technique to achieve parasitic transient cancellation, and FPGA-based data acquisition firmware was implemented to provide an efficient, flexible multiple-gate sequencing methodology for obtaining the dependence
Zhonghua Nan Ke Xue National Journal of Andrology, Oct 1, 2009
To establish a primary culture of the testis gubernacular cells of Kunming mice, observe the morp... more To establish a primary culture of the testis gubernacular cells of Kunming mice, observe the morphological characteristics of the cells, and explore the effects of exogenous estrogens (EEs) on the development of the testis gubernacula in vitro. We removed the gubernacula from 3-day-old mice with the surgical magnifier and cultured the gubernacular cells. Then we detected the cell viability by trypan blue and cell morphology by HE staining. The subcultured cells were randomly divided into a blank control, a DMSO (0.1%, v/v) control, and 4 experimental groups (given 0.01, 0.10, 1.00 and 10.00 micdrog/ml of diethylstilbestrol [DES] dissolved in DMSO, respectively). After treated for 12, 24 and 48 hours, the gubernacular cells were observed for morphological changes and proliferation inhibition by CCK-8. Most of the cultured gubernacular cells were fibroblasts, and a few were epithelioids. The primary cells showed a viability of 85%-90%. Dose- and time-dependent inhibition of cell proliferation was found in the four experimental groups at three different times, with statistically significant differences (P < 0.01). Gubernacular cells can be cultured in vitro. EEs inhibit the proliferation of gubernacular cells in a dose- and time-dependent manner. An in- sight into the effects EES on cultured gubernacular cells is an effective approach to the study of their influence on the development of the reproductive system.
Deep learning models are shown to be vulnerable to adversarial examples. Though adversarial train... more Deep learning models are shown to be vulnerable to adversarial examples. Though adversarial training can enhance model robustness, typical approaches are computationally expensive. Recent works proposed to transfer the robustness to adversarial attacks across different tasks or models with soft labels. Compared to soft labels, feature contains rich semantic information and holds the potential to be applied to different downstream tasks. In this paper, we propose a novel approach called Guided Adversarial Contrastive Distillation (GACD), to effectively transfer adversarial robustness from teacher to student with features. We first formulate this objective as contrastive learning and connect it with mutual information. With a well-trained teacher model as an anchor, students are expected to extract features similar to the teacher. Then considering the potential errors made by teachers, we propose sample reweighted estimation to eliminate the negative effects from teachers. With GACD, the student not only learns to extract robust features, but also captures structural knowledge from the teacher. By extensive experiments evaluating over popular datasets such as CIFAR-10, CIFAR-100 and STL-10, we demonstrate that our approach can effectively transfer robustness across different models and even different tasks, and achieve comparable or better results than existing methods. Besides, we provide a detailed analysis of various methods, showing that students produced by our approach capture more structural knowledge from teachers and learn more robust features under adversarial attacks.
Large-scale, highly integrated and low-power-consuming hardware is becoming progressively more im... more Large-scale, highly integrated and low-power-consuming hardware is becoming progressively more important for realizing optical neural networks (ONNs) capable of advanced optical computing. Traditional experimental implementations need N2 units such as Mach-Zehnder interferometers (MZIs) for an input dimension N to realize typical computing operations (convolutions and matrix multiplication), resulting in limited scalability and consuming excessive power. Here, we propose the integrated diffractive optical network for implementing parallel Fourier transforms, convolution operations and application-specific optical computing using two ultracompact diffractive cells (Fourier transform operation) and only N MZIs. The footprint and energy consumption scales linearly with the input data dimension, instead of the quadratic scaling in the traditional ONN framework. A ~10-fold reduction in both footprint and energy consumption, as well as equal high accuracy with previous MZI-based ONNs was ...
2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021
This paper studies the problem of panoramic image reflection removal, aiming at reliving the cont... more This paper studies the problem of panoramic image reflection removal, aiming at reliving the content ambiguity between reflection and transmission scenes. Although a partial view of the reflection scene is included in the panoramic image, it cannot be utilized directly due to its misalignment with the reflection-contaminated image. We propose a two-step approach to solve this problem, by first accomplishing geometric and photometric alignment for the reflection scene via a coarse-to-fine strategy, and then restoring the transmission scene via a recovery network. The proposed method is trained with a synthetic dataset and verified quantitatively with a real panoramic image dataset. The effectiveness of the proposed method is validated by the significant performance advantage over single image-based reflection removal methods and generalization capacity to limited-FoV scenarios captured by conventional camera or mobile phone users
Complex-valued neural networks have many advantages over their real-valued counterparts. Conventi... more Complex-valued neural networks have many advantages over their real-valued counterparts. Conventional digital electronic computing platforms are incapable of executing truly complex-valued representations and operations. In contrast, optical computing platforms that encode information in both phase and magnitude can execute complex arithmetic by optical interference, offering significantly enhanced computational speed and energy efficiency. However, to date, most demonstrations of optical neural networks still only utilize conventional real-valued frameworks that are designed for digital computers, forfeiting many of the advantages of optical computing such as efficient complex-valued operations. In this article, we highlight an optical neural chip (ONC) that implements truly complex-valued neural networks. We benchmark the performance of our complex-valued ONC in four settings: simple Boolean tasks, species classification of an Iris dataset, classifying nonlinear datasets (Circle a...
Gait recognition is an emerging biometric technology that identifies people through the analysis ... more Gait recognition is an emerging biometric technology that identifies people through the analysis of the way they walk. The challenge of model-free based gait recognition is to cope with various intra-class variations such as clothing variations, carrying conditions and angle variations that adversely affect the recognition performance. This paper proposes a method to select the most discriminative human body part based on group Lasso of motion to reduce the intra-class variation so as to improve the recognition performance. The proposed method is evaluated using CASIA Gait Dataset B. Experimental results demonstrate that the proposed technique gives promising results.
Semantic image segmentation aims to classify every pixel of a scene image to one of many classes.... more Semantic image segmentation aims to classify every pixel of a scene image to one of many classes. It implicitly involves object recognition, localization, and boundary delineation. In this paper, we propose a segmentation network called CGBNet to enhance the segmentation performance by context encoding and multi-path decoding. We first propose a context encoding module that generates context-contrasted local feature to make use of the informative context and the discriminative local information. This context encoding module greatly improves the segmentation performance, especially for inconspicuous objects. Furthermore, we propose a scale-selection scheme to selectively fuse the segmentation results from different-scales of features at every spatial position. It adaptively selects appropriate score maps from rich scales of features. To improve the segmentation performance results at boundary, we further propose a boundary delineation module that encourages the location-specific verylow-level features near the boundaries to take part in the final prediction and suppresses them far from the boundaries. The proposed segmentation network achieves very competitive performance in terms of all three different evaluation metrics consistently on the six popular scene segmentation datasets, Pascal Context, SUN-RGBD, Sift Flow, COCO Stuff, ADE20K, and Cityscapes.
• Identification requires comparison of a person’s fingerprint with all fingerprints in a fingerp... more • Identification requires comparison of a person’s fingerprint with all fingerprints in a fingerprint database. Most of existing fingerprint datasets are very large (>1 million fingerprints). • In some cases other attributes such as race, gender, age, soft biometrics are introduces in the database, which allows partitioning of the dataset into smaller subsets. However, in most cases the data are fingerprints only. • Comparison of an on-line acquired fingerprint with all fingerprints in the database is computationally expensive (FBI dataset has >200 million fingerprints). • A solution to this problem is to divide the database into a number of bins (based on some predefined automatically extracted general fingerprint features). • Classification is referred to the assigning a class in consistent and reliable way. • Fingerprint matching is based on local features while fingerprint classification is based on global features, such as global ridge patter and singularities.
ABSTRACT Due to high dimensionality of images or generated color features, different color channe... more ABSTRACT Due to high dimensionality of images or generated color features, different color channels are usually processed separately and then concatenated together into a feature vector for classification. This makes channel fusion a crucial step in color face recognition (FR) systems. However, existing methods simply concatenate channel-wise color features without identifying the importance or reliability of features in different color channels. In this paper, we propose a color channel fusion (CCF) approach using jointly dimension reduction algorithms to select more features from reliable and discriminative channels. Experiments using two different dimension reduction approaches, two different types of features on three image datasets show that CCF achieves consistently better performance than color channel concatenation (CCC) method which deals with different color channels equally.
In this paper, we address the false rejection problem due to small solid state sensor area availa... more In this paper, we address the false rejection problem due to small solid state sensor area available for fingerprint image capture. We propose a minutiae data synthesis approach to circumvent this problem. Main advantages of this approach over existing image mosaicing approach include low memory storage requirement and low computational complexity. Moreover, possible overhead on the search engine (for fingerprint matching) due to data redundancy could be reduced. Extensive experiments were conducted to determine the best transformation suitable for minutiae alignment. We demonstrate the idea of synthesis with an example using physical fingerprint images. The proposed synthesis system is also found to improve (lower) the number of false rejects due to the use of different fingerprint regions for matching.
This paper proposes a novel full-reference quality assessment (QA) metric that automatically asse... more This paper proposes a novel full-reference quality assessment (QA) metric that automatically assesses the quality of an image in the discrete orthogonal moments domain. This metric is constructed by representing the spatial information of an image using low order moments. The computation, up to fourth order moments, is performed on each individual ð8 Â 8Þ non-overlapping block for both the test and reference images. Then, the computed moments of both the test and reference images are combined in order to determine the moment correlation index of each block in each order. The number of moment correlation indices used in this study is nine. Next, the mean of each moment correlation index is computed and thereafter the single quality interpretation of the test image with respect to its reference is determined by taking the mean value of the computed means of all the moment correlation indices. The proposed objective metrics based on two discrete orthogonal moments, Tchebichef and Krawtchouk moments, are developed and their performances are evaluated by comparing them with subjective ratings on several publicly available databases. The proposed discrete orthogonal moments based metric performs competitively well with the state-of-the-art models in terms of quality prediction while outperforms them in terms of computational speed.
Discussion papers are research materials circulated by their authors for purposes of information ... more Discussion papers are research materials circulated by their authors for purposes of information and discussion. They have not necessarily undergone formal peer review.
ABSTRACT JIANG X., DIETZENBACHER E. and LOS B. Improved estimation of regional input-output table... more ABSTRACT JIANG X., DIETZENBACHER E. and LOS B. Improved estimation of regional input-output tables using cross-regional methods, Regional Studies. Many regional input-output tables are estimated by means of non-survey methods. Often, information on the margins of the projected table is complemented by full information on intermediate inputs from tables for other regions. This paper compares the performance of four of such 'cross-regional' methods. Two of these were already proposed in the literature, whereas the other two are based on recent advances in regression analysis. The methods are tested not only against each other, but also against traditional methods that do not employ cross-regional information. To this end, twenty-seven regional input-output tables for China in 1997 and 2002 are used.
Beijing Da Xue Xue Bao Yi Xue Ban Journal of Peking University Health Sciences, Jul 1, 2003
To investigate the feasibility and cortex activation of fMRI in aged volunteers during the perfor... more To investigate the feasibility and cortex activation of fMRI in aged volunteers during the performance of calculation tasks. Internal mental calculation tasks including simple tasks and complicated tasks were conducted in 11 normal aged volunteers. The fMRI data was postprocessed using SPM99 to reveal the activated cortex. The simple calculation tasks were fulfilled satisfactorily in all aged subjects. Cortex in the bilateral Superior parietal lobule, inferior parietal lobule and bilateral occipital lobes showed activation. We found a selective enhancement of fMRI signal in Brodmann regions 9, 10 and 46 in the complicated tasks. fMRI tasks are feasible in aged people and more cortex of the left frontal lobe shows activation in complicated calculation tasks.
Http Dx Doi Org 10 1080 09500340 2012 698659, Oct 10, 2012
The characterization and analysis of afterpulsing behavior in InGaAs/InP single photon avalanche ... more The characterization and analysis of afterpulsing behavior in InGaAs/InP single photon avalanche diodes (SPADs) is reported for gating frequencies between 10 and 50 MHz. Gating in this frequency range was accomplished using a matched delay line technique to achieve parasitic transient cancellation, and FPGA-based data acquisition firmware was implemented to provide an efficient, flexible multiple-gate sequencing methodology for obtaining the dependence
Zhonghua Nan Ke Xue National Journal of Andrology, Oct 1, 2009
To establish a primary culture of the testis gubernacular cells of Kunming mice, observe the morp... more To establish a primary culture of the testis gubernacular cells of Kunming mice, observe the morphological characteristics of the cells, and explore the effects of exogenous estrogens (EEs) on the development of the testis gubernacula in vitro. We removed the gubernacula from 3-day-old mice with the surgical magnifier and cultured the gubernacular cells. Then we detected the cell viability by trypan blue and cell morphology by HE staining. The subcultured cells were randomly divided into a blank control, a DMSO (0.1%, v/v) control, and 4 experimental groups (given 0.01, 0.10, 1.00 and 10.00 micdrog/ml of diethylstilbestrol [DES] dissolved in DMSO, respectively). After treated for 12, 24 and 48 hours, the gubernacular cells were observed for morphological changes and proliferation inhibition by CCK-8. Most of the cultured gubernacular cells were fibroblasts, and a few were epithelioids. The primary cells showed a viability of 85%-90%. Dose- and time-dependent inhibition of cell proliferation was found in the four experimental groups at three different times, with statistically significant differences (P < 0.01). Gubernacular cells can be cultured in vitro. EEs inhibit the proliferation of gubernacular cells in a dose- and time-dependent manner. An in- sight into the effects EES on cultured gubernacular cells is an effective approach to the study of their influence on the development of the reproductive system.
Uploads
Papers by Xudong Jiang