Mobile devices have access to personal, potentially sensitive data, and there is a growing number... more Mobile devices have access to personal, potentially sensitive data, and there is a growing number of applications that transmit this personally identifiable information (PII) over the network. In this paper, we present the AntShield system that performs on-device packet-level monitoring and detects the transmission of such sensitive information accurately and in real-time. A key insight is to distinguish PII that is predefined and is easily available on the device from PII that is unknown a priori but can be automatically detected by classifiers. Our system not only combines, for the first time, the advantages of on-device monitoring with the power of learning unknown PII, but also outperforms either of the two approaches alone. We demonstrate the real-time performance of our prototype as well as the classification performance using a dataset that we collect and analyze from scratch (including new findings in terms of leaks and patterns). AntShield is a first step towards enabling distributed learning of private information exposure.
2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
Few-shot learning (FSL) aims to learn new categories with a few visual samples per class. Few-sho... more Few-shot learning (FSL) aims to learn new categories with a few visual samples per class. Few-shot class representations are often biased due to data scarcity. To mitigate this issue, we propose to generate visual samples based on semantic embeddings using a conditional variational autoencoder (CVAE) model. We train this CVAE model on base classes and use it to generate features for novel classes. More importantly, we guide this VAE to strictly generate representative samples by removing nonrepresentative samples from the base training set when training the CVAE model. We show that this training scheme enhances the representativeness of the generated samples and therefore, improves the few-shot classification results. Experimental results show that our method improves three FSL baseline methods by substantial margins, achieving state-of-the-art few-shot classification performance on miniImageNet and tieredImageNet datasets for both 1-shot and 5-shot settings. Code is available at: https://github.com/cvlab-stonybrook/ fsl-rsvae.
2017 IEEE 18th International Symposium on A World of Wireless, Mobile and Multimedia Networks (WoWMoM), 2017
The increasing number of mobile devices with high processing power and high-resolution screens ha... more The increasing number of mobile devices with high processing power and high-resolution screens had led to an enormous growth of mobile video traffic. Mobile network operators face the requirement to efficiently support large numbers of concurrent unicast streaming sessions. In the present work, the long-term quality of experience perceived by the user, the fairness, and the overall system efficiency are addressed simultaneously from the cross-layer perspective by jointly optimizing the video adaptation and the wireless resource allocation. One fundamental challenge of the cross layer design is that the time scale of video adaptation-seconds-differs by several orders of magnitude from the one of resource allocation-milliseconds. We focus on the low-delay live streaming, which is particularly sensitive to the throughput fluctuation. We consider the streaming both in the downlink and in the uplink, explicitly taking into account the imperfect synchronization in the uplink. Our proposed solution consists of two components. First, we formulate the problem of video adaptation as a quality of experience based maxmin optimization problem that leverages the link rate estimate in the lower network layers. Second, we propose a dynamic resource allocation scheme that takes into account the demands of the streaming clients. These two components together aim at a fair and efficient cross-layer streaming system. An accurate estimation of link rate on the time scale of seconds, required for this problem, is particularly difficult in mobile networks. As a separate contribution, several link rate estimation approaches are evaluated. The prediction algorithms assuming the static resource allocation, although computationally less complex, may lead to inaccurate prediction results in comparison to the one when dynamic resource allocation scheme is used. In this work, the spectral efficiency gain by dynamic resource allocation can be approximated and used to improve throughput predictions. We evaluate the proposed approach against state-of-the-art baselines. The results reveal significant improvements of quality of experience in all studied use cases.
Epigenetic alterations found in all human cancers are promising targets for anticancer therapy. I... more Epigenetic alterations found in all human cancers are promising targets for anticancer therapy. In this sense, histone deacetylase inhibitors (HDACIs) are interesting anticancer agents that play an important role in the epigenetic regulation of cancer cells. Here, we report 15 novel hydroxamic acid-based histone deacetylase inhibitors with quinazolinone core structures. Five compounds exhibited antiproliferative activity with IC50 values of 3.4–37.8 µM. Compound 8 with a 2-mercaptoquinazolinone cap moiety displayed the highest antiproliferative efficacy against MCF-7 cells. For the HDAC6 target selectivity study, compound 8 displayed an IC50 value of 2.3 µM, which is 29.3 times higher than those of HDAC3, HDAC4, HDAC8, and HDAC11. Western blot assay proved that compound 8 strongly inhibited tubulin acetylation, a substrate of HDAC6. Compound 8 also displayed stronger inhibition activity against HDAC11 than the control drug Belinostat. The inhibitory mechanism of action of compound 8...
The requirement for paired shadow and shadow-free images limits the size and diversity of shadow ... more The requirement for paired shadow and shadow-free images limits the size and diversity of shadow removal datasets and hinders the possibility of training large-scale, robust shadow removal algorithms. We propose a shadow removal method that can be trained using only shadow and non-shadow patches cropped from the shadow images themselves. Our method is trained via an adversarial framework, following a physical model of shadow formation. Our central contribution is a set of physics-based constraints that enables this adversarial training. Our method achieves competitive shadow removal results compared to state-of-the-art methods that are trained with fully paired shadow and shadow-free images. The advantages of our training regime are even more pronounced in shadow removal for videos. Our method can be fine-tuned on a testing video with only the shadow masks generated by a pre-trained shadow detector and outperforms state-of-the-art methods on this challenging test. We illustrate the advantages of our method on our proposed video shadow removal dataset.
2021 IEEE/CVF International Conference on Computer Vision (ICCV), 2021
Data augmentation is an intuitive step towards solving the problem of few-shot classification. Ho... more Data augmentation is an intuitive step towards solving the problem of few-shot classification. However, ensuring both discriminability and diversity in the augmented samples is challenging. To address this, we propose a feature disentanglement framework that allows us to augment features with randomly sampled intra-class variations while preserving their class-discriminative features. Specifically, we disentangle a feature representation into two components: one represents the intra-class variance and the other encodes the class-discriminative information. We assume that the intra-class variance induced by variations in poses, backgrounds, or illumination conditions is shared across all classes and can be modelled via a common distribution. Then we sample features repeatedly from the learned intraclass variability distribution and add them to the classdiscriminative features to get the augmented features. Such a data augmentation scheme ensures that the augmented features inherit crucial class-discriminative features while exhibiting large intra-class variance. Our method significantly outperforms the state-of-the-art methods on multiple challenging fine-grained few-shot image classification benchmarks.
The adblocking arms race has escalated over the last few years. An entire new ecosystem of circum... more The adblocking arms race has escalated over the last few years. An entire new ecosystem of circumvention (CV) services has recently emerged that aims to bypass adblockers by obfuscating site content, making it difficult for adblocking filter lists to distinguish between ads and functional content. In this paper, we investigate recent anti-circumvention efforts by the adblocking community that leverage custom filter lists. In particular, we analyze the anti-circumvention filter list (ACVL), which supports advanced filter rules with enriched syntax and capabilities designed specifically to counter circumvention. We show that keeping ACVL rules up-to-date requires expert list curators to continuously monitor sites known to employ CV services and to discover new such sites in the wild — both tasks require considerable manual effort. To help automate and scale ACVL curation, we develop CV-INSPECTOR, a machine learning approach for automatically detecting adblock circumvention using diffe...
In the uplink of OFDMA-based systems, Multiple Access Interference (MAI), caused by synchronizati... more In the uplink of OFDMA-based systems, Multiple Access Interference (MAI), caused by synchronization offsets, can cause considerable degradation in the performance of the User Terminals (UTs). The conventional approach to deal with MAI is based on the usage of Cyclic Prefix (CP) also known as Guard Intervals (GI) in time domain. The time and frequency offsets of UTs signals arriving at BS are estimated and then these estimations are used to subtract the offsets and to re-build orthogonality among subcarriers. However the large overhead, caused by long CP and pilot subcarriers, and imposed by this approach, is considered as a main drawback of this approach. Another way of mitigating MAI is by inserting frequency Guard Bands (GB) to reduce the MAI on adjacent subcarriers. In this report, we examine the feasibility of using GB together with CP instead of only CP using a simple standard model. We consider a scenario where a fixed-width CP is used with fixed-width GB to mitigate MAI and i...
Motivated by the growing popularity of smart TVs, we present a large-scale measurement study of s... more Motivated by the growing popularity of smart TVs, we present a large-scale measurement study of smart TVs by collecting and analyzing their network traffic from two different vantage points. First, we analyze aggregate network traffic of smart TVs in-the-wild, collected from residential gateways of tens of homes and several different smart TV platforms, including Apple, Samsung, Roku, and Chromecast. In addition to accessing video streaming and cloud services, we find that smart TVs frequently connect to well-known as well as platform-specific advertising and tracking services (ATS). Second, we instrument Roku and Amazon Fire TV, two popular smart TV platforms, by setting up a controlled testbed to systematically exercise the top-1000 apps on each platform, and analyze their network traffic at the granularity of the individual apps. We again find that smart TV apps connect to a wide range of ATS, and that the key players of the ATS ecosystems of the two platforms are different from ...
This paper proposes a geodesic-distance-based feature that encodes global information for improve... more This paper proposes a geodesic-distance-based feature that encodes global information for improved video segmentation algorithms. The feature is a joint histogram of intensity and geodesic distances, where the geodesic distances are computed as the shortest paths between superpixels via their boundaries. We also incorporate adaptive voting weights and spatial pyramid configurations to include spatial information into the geodesic histogram feature and show that this further improves results. The feature is generic and can be used as part of various algorithms. In experiments, we test the geodesic histogram feature by incorporating it into two existing video segmentation frameworks. This leads to significantly better performance in 3D video segmentation benchmarks on two datasets.
We propose a novel GAN-based framework for detecting shadows in images, in which a shadow detecti... more We propose a novel GAN-based framework for detecting shadows in images, in which a shadow detection network (D-Net) is trained together with a shadow attenuation network (A-Net) that generates adversarial training examples. The A-Net modifies the original training images constrained by a simplified physical shadow model and is focused on fooling the D-Net’s shadow predictions. Hence, it is effectively augmenting the training data for D-Net with hard-to-predict cases. The D-Net is trained to predict shadows in both original images and generated images from the A-Net. Our experimental results show that the additional training data from A-Net significantly improves the shadow detection accuracy of D-Net. Our method outperforms the stateof-the-art methods on the most challenging shadow detection benchmark (SBU) and also obtains state-of-the-art results on a cross-dataset task, testing on UCF. Furthermore, the proposed method achieves accurate real-time shadow detection at 45 frames per ...
IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021
We propose a novel deep learning method for shadow removal. Inspired by physical models of shadow... more We propose a novel deep learning method for shadow removal. Inspired by physical models of shadow formation, we use a linear illumination transformation to model the shadow effects in the image that allows the shadow image to be expressed as a combination of the shadow-free image, the shadow parameters, and a matte layer. We use two deep networks, namely SP-Net and M-Net, to predict the shadow parameters and the shadow matte respectively. This system allows us to remove the shadow effects from images. We then employ an inpainting network, I-Net, to further refine the results. We train and test our framework on the most challenging shadow removal dataset (ISTD). Our method improves the state-of-the-art in terms of mean absolute error (MAE) for the shadow area by 20%. Furthermore, this decomposition allows us to formulate a patch-based weakly-supervised shadow removal method. This model can be trained without any shadow-free images (that are cumbersome to acquire) and achieves competitive shadow removal results compared to state-of-the-art methods that are trained with fully paired shadow and shadow-free images. Last, we introduce SBU-Timelapse, a video shadow removal dataset for evaluating shadow removal methods.
We propose a novel GAN-based framework for detecting shadows in images, in which a shadow detecti... more We propose a novel GAN-based framework for detecting shadows in images, in which a shadow detection network (D-Net) is trained together with a shadow attenuation network (A-Net) that generates adversarial training examples. The A-Net modifies the original training images constrained by a simplified physical shadow model and is focused on fooling the D-Net's shadow predictions. Hence, it is effectively augmenting the training data for D-Net with hard-to-predict cases. The D-Net is trained to predict shadows in both original images and generated images from the A-Net. Our experimental results show that the additional training data from A-Net significantly improves the shadow detection accuracy of D-Net. Our method outperforms the stateof-the-art methods on the most challenging shadow detection benchmark (SBU) and also obtains state-of-the-art results on a cross-dataset task, testing on UCF. Furthermore, the proposed method achieves accurate real-time shadow detection at 45 frames per second.
Proceedings on Privacy Enhancing Technologies, 2020
In this paper, we present a large-scale measurement study of the smart TV advertising and trackin... more In this paper, we present a large-scale measurement study of the smart TV advertising and tracking ecosystem. First, we illuminate the network behavior of smart TVs as used in the wild by analyzing network traffic collected from residential gateways. We find that smart TVs connect to well-known and platform-specific advertising and tracking services (ATSes). Second, we design and implement software tools that systematically explore and collect traffic from the top-1000 apps on two popular smart TV platforms, Roku and Amazon Fire TV. We discover that a subset of apps communicate with a large number of ATSes, and that some ATS organizations only appear on certain platforms, showing a possible segmentation of the smart TV ATS ecosystem across platforms. Third, we evaluate the (in)effectiveness of DNS-based blocklists in preventing smart TVs from accessing ATSes. We highlight that even smart TV-specific blocklists suffer from missed ads and incur functionality breakage. Finally, we exam...
2017 IEEE International Conference on Computer Vision Workshops (ICCVW), 2017
Co-localization is the problem of localizing categorical objects using only positive set of examp... more Co-localization is the problem of localizing categorical objects using only positive set of example images, without any form of further supervision. This is a challenging task as there is no pixel-level annotations. Motivated by human visual learning, we find the common features of an object category from convolutional kernels of a pretrained convolutional neural network (CNN). We call these category-consistent CNN features. Then, we co-propagate their activated spatial regions using superpixel geodesic distances for localization. In our first set of experiments, we show that the proposed method achieves state-of-the-art performance on three related benchmarks: PASCAL 2007, PASCAL-2012, and the Object Discovery dataset. We also show that our method is able to detect and localize truly unseen categories, using six held-out ImagNet subset of categories with state-of-the-art accuracies. Our intuitive approach achieves this success without any region proposals or object detectors, and can be based on a CNN that was pre-trained purely on image classification tasks without further fine-tuning.
In this work, we tackle the problem of crowd counting in images. We present a Convolutional Neura... more In this work, we tackle the problem of crowd counting in images. We present a Convolutional Neural Network (CNN) based density estimation approach to solve this problem. Predicting a high resolution density map in one go is a challenging task. Hence, we present a two branch CNN architecture for generating high resolution density maps, where the first branch generates a low resolution density map, and the second branch incorporates the low resolution prediction and feature maps from the first branch to generate a high resolution density map. We also propose a multi-stage extension of our approach where each stage in the pipeline utilizes the predictions from all the previous stages. Empirical comparison with the previous state-of-the-art crowd counting methods shows that our method achieves the lowest mean absolute error on three challenging crowd counting benchmarks: Shanghaitech, World-Expo'10, and UCF datasets.
2019 IEEE/CVF International Conference on Computer Vision (ICCV), 2019
We propose a novel deep learning method for shadow removal. Inspired by physical models of shadow... more We propose a novel deep learning method for shadow removal. Inspired by physical models of shadow formation, we use a linear illumination transformation to model the shadow effects in the image that allows the shadow image to be expressed as a combination of the shadow-free image, the shadow parameters, and a matte layer. We use two deep networks, namely SP-Net and M-Net, to predict the shadow parameters and the shadow matte respectively. This system allows us to remove the shadow effects on the images. We train and test our framework on the most challenging shadow removal dataset (ISTD). Compared to the state-ofthe-art method, our model achieves a 40% error reduction in terms of root mean square error (RMSE) for the shadow area, reducing RMSE from 13.3 to 7.9. Moreover, we create an augmented ISTD dataset based on an image decomposition system by modifying the shadow parameters to generate new synthetic shadow images. Training our model on this new augmented ISTD dataset further lowers the RMSE on the shadow area to 7.4.
Nowadays, deep learning is becoming increasingly important in our daily life. The appearance of d... more Nowadays, deep learning is becoming increasingly important in our daily life. The appearance of deep learning in many applications in life relates to prediction and classification such as self-driving, product recommendation, advertisements and healthcare. Therefore, if a deep learning model causes false predictions and misclassification, it can do great harm. This is basically a crucial issue in the deep learning model. In addition, deep learning models use large amounts of data in the training/learning phases, which contain sensitive information. Therefore, when deep learning models are used in real-world applications, it is required to protect the privacy information used in the model. In this article, we carry out a brief review of the threats and defenses methods on security issues for the deep learning models and the privacy of the data used in such models while maintaining their performance and accuracy. Finally, we discuss current challenges and future developments.
Proceedings of the Fifth International Conference on Web Information Systems and Technologies, 2009
This paper studies the problem of classifying structured data sources on the Web. While prior wor... more This paper studies the problem of classifying structured data sources on the Web. While prior works use all features, once extracted from search interfaces, we further refine the feature set. In our research, each search interface is treated simply as a bag-of-words. We choose a subset of words, which is suited to classify web sources, by our feature selection methods with new metrics and a novel simple ranking scheme. Using aggressive feature selection approach, together with a Gaussian process classifier, we obtained high classification performance in an evaluation over real web data.
Mobile devices have access to personal, potentially sensitive data, and there is a growing number... more Mobile devices have access to personal, potentially sensitive data, and there is a growing number of applications that transmit this personally identifiable information (PII) over the network. In this paper, we present the AntShield system that performs on-device packet-level monitoring and detects the transmission of such sensitive information accurately and in real-time. A key insight is to distinguish PII that is predefined and is easily available on the device from PII that is unknown a priori but can be automatically detected by classifiers. Our system not only combines, for the first time, the advantages of on-device monitoring with the power of learning unknown PII, but also outperforms either of the two approaches alone. We demonstrate the real-time performance of our prototype as well as the classification performance using a dataset that we collect and analyze from scratch (including new findings in terms of leaks and patterns). AntShield is a first step towards enabling distributed learning of private information exposure.
2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
Few-shot learning (FSL) aims to learn new categories with a few visual samples per class. Few-sho... more Few-shot learning (FSL) aims to learn new categories with a few visual samples per class. Few-shot class representations are often biased due to data scarcity. To mitigate this issue, we propose to generate visual samples based on semantic embeddings using a conditional variational autoencoder (CVAE) model. We train this CVAE model on base classes and use it to generate features for novel classes. More importantly, we guide this VAE to strictly generate representative samples by removing nonrepresentative samples from the base training set when training the CVAE model. We show that this training scheme enhances the representativeness of the generated samples and therefore, improves the few-shot classification results. Experimental results show that our method improves three FSL baseline methods by substantial margins, achieving state-of-the-art few-shot classification performance on miniImageNet and tieredImageNet datasets for both 1-shot and 5-shot settings. Code is available at: https://github.com/cvlab-stonybrook/ fsl-rsvae.
2017 IEEE 18th International Symposium on A World of Wireless, Mobile and Multimedia Networks (WoWMoM), 2017
The increasing number of mobile devices with high processing power and high-resolution screens ha... more The increasing number of mobile devices with high processing power and high-resolution screens had led to an enormous growth of mobile video traffic. Mobile network operators face the requirement to efficiently support large numbers of concurrent unicast streaming sessions. In the present work, the long-term quality of experience perceived by the user, the fairness, and the overall system efficiency are addressed simultaneously from the cross-layer perspective by jointly optimizing the video adaptation and the wireless resource allocation. One fundamental challenge of the cross layer design is that the time scale of video adaptation-seconds-differs by several orders of magnitude from the one of resource allocation-milliseconds. We focus on the low-delay live streaming, which is particularly sensitive to the throughput fluctuation. We consider the streaming both in the downlink and in the uplink, explicitly taking into account the imperfect synchronization in the uplink. Our proposed solution consists of two components. First, we formulate the problem of video adaptation as a quality of experience based maxmin optimization problem that leverages the link rate estimate in the lower network layers. Second, we propose a dynamic resource allocation scheme that takes into account the demands of the streaming clients. These two components together aim at a fair and efficient cross-layer streaming system. An accurate estimation of link rate on the time scale of seconds, required for this problem, is particularly difficult in mobile networks. As a separate contribution, several link rate estimation approaches are evaluated. The prediction algorithms assuming the static resource allocation, although computationally less complex, may lead to inaccurate prediction results in comparison to the one when dynamic resource allocation scheme is used. In this work, the spectral efficiency gain by dynamic resource allocation can be approximated and used to improve throughput predictions. We evaluate the proposed approach against state-of-the-art baselines. The results reveal significant improvements of quality of experience in all studied use cases.
Epigenetic alterations found in all human cancers are promising targets for anticancer therapy. I... more Epigenetic alterations found in all human cancers are promising targets for anticancer therapy. In this sense, histone deacetylase inhibitors (HDACIs) are interesting anticancer agents that play an important role in the epigenetic regulation of cancer cells. Here, we report 15 novel hydroxamic acid-based histone deacetylase inhibitors with quinazolinone core structures. Five compounds exhibited antiproliferative activity with IC50 values of 3.4–37.8 µM. Compound 8 with a 2-mercaptoquinazolinone cap moiety displayed the highest antiproliferative efficacy against MCF-7 cells. For the HDAC6 target selectivity study, compound 8 displayed an IC50 value of 2.3 µM, which is 29.3 times higher than those of HDAC3, HDAC4, HDAC8, and HDAC11. Western blot assay proved that compound 8 strongly inhibited tubulin acetylation, a substrate of HDAC6. Compound 8 also displayed stronger inhibition activity against HDAC11 than the control drug Belinostat. The inhibitory mechanism of action of compound 8...
The requirement for paired shadow and shadow-free images limits the size and diversity of shadow ... more The requirement for paired shadow and shadow-free images limits the size and diversity of shadow removal datasets and hinders the possibility of training large-scale, robust shadow removal algorithms. We propose a shadow removal method that can be trained using only shadow and non-shadow patches cropped from the shadow images themselves. Our method is trained via an adversarial framework, following a physical model of shadow formation. Our central contribution is a set of physics-based constraints that enables this adversarial training. Our method achieves competitive shadow removal results compared to state-of-the-art methods that are trained with fully paired shadow and shadow-free images. The advantages of our training regime are even more pronounced in shadow removal for videos. Our method can be fine-tuned on a testing video with only the shadow masks generated by a pre-trained shadow detector and outperforms state-of-the-art methods on this challenging test. We illustrate the advantages of our method on our proposed video shadow removal dataset.
2021 IEEE/CVF International Conference on Computer Vision (ICCV), 2021
Data augmentation is an intuitive step towards solving the problem of few-shot classification. Ho... more Data augmentation is an intuitive step towards solving the problem of few-shot classification. However, ensuring both discriminability and diversity in the augmented samples is challenging. To address this, we propose a feature disentanglement framework that allows us to augment features with randomly sampled intra-class variations while preserving their class-discriminative features. Specifically, we disentangle a feature representation into two components: one represents the intra-class variance and the other encodes the class-discriminative information. We assume that the intra-class variance induced by variations in poses, backgrounds, or illumination conditions is shared across all classes and can be modelled via a common distribution. Then we sample features repeatedly from the learned intraclass variability distribution and add them to the classdiscriminative features to get the augmented features. Such a data augmentation scheme ensures that the augmented features inherit crucial class-discriminative features while exhibiting large intra-class variance. Our method significantly outperforms the state-of-the-art methods on multiple challenging fine-grained few-shot image classification benchmarks.
The adblocking arms race has escalated over the last few years. An entire new ecosystem of circum... more The adblocking arms race has escalated over the last few years. An entire new ecosystem of circumvention (CV) services has recently emerged that aims to bypass adblockers by obfuscating site content, making it difficult for adblocking filter lists to distinguish between ads and functional content. In this paper, we investigate recent anti-circumvention efforts by the adblocking community that leverage custom filter lists. In particular, we analyze the anti-circumvention filter list (ACVL), which supports advanced filter rules with enriched syntax and capabilities designed specifically to counter circumvention. We show that keeping ACVL rules up-to-date requires expert list curators to continuously monitor sites known to employ CV services and to discover new such sites in the wild — both tasks require considerable manual effort. To help automate and scale ACVL curation, we develop CV-INSPECTOR, a machine learning approach for automatically detecting adblock circumvention using diffe...
In the uplink of OFDMA-based systems, Multiple Access Interference (MAI), caused by synchronizati... more In the uplink of OFDMA-based systems, Multiple Access Interference (MAI), caused by synchronization offsets, can cause considerable degradation in the performance of the User Terminals (UTs). The conventional approach to deal with MAI is based on the usage of Cyclic Prefix (CP) also known as Guard Intervals (GI) in time domain. The time and frequency offsets of UTs signals arriving at BS are estimated and then these estimations are used to subtract the offsets and to re-build orthogonality among subcarriers. However the large overhead, caused by long CP and pilot subcarriers, and imposed by this approach, is considered as a main drawback of this approach. Another way of mitigating MAI is by inserting frequency Guard Bands (GB) to reduce the MAI on adjacent subcarriers. In this report, we examine the feasibility of using GB together with CP instead of only CP using a simple standard model. We consider a scenario where a fixed-width CP is used with fixed-width GB to mitigate MAI and i...
Motivated by the growing popularity of smart TVs, we present a large-scale measurement study of s... more Motivated by the growing popularity of smart TVs, we present a large-scale measurement study of smart TVs by collecting and analyzing their network traffic from two different vantage points. First, we analyze aggregate network traffic of smart TVs in-the-wild, collected from residential gateways of tens of homes and several different smart TV platforms, including Apple, Samsung, Roku, and Chromecast. In addition to accessing video streaming and cloud services, we find that smart TVs frequently connect to well-known as well as platform-specific advertising and tracking services (ATS). Second, we instrument Roku and Amazon Fire TV, two popular smart TV platforms, by setting up a controlled testbed to systematically exercise the top-1000 apps on each platform, and analyze their network traffic at the granularity of the individual apps. We again find that smart TV apps connect to a wide range of ATS, and that the key players of the ATS ecosystems of the two platforms are different from ...
This paper proposes a geodesic-distance-based feature that encodes global information for improve... more This paper proposes a geodesic-distance-based feature that encodes global information for improved video segmentation algorithms. The feature is a joint histogram of intensity and geodesic distances, where the geodesic distances are computed as the shortest paths between superpixels via their boundaries. We also incorporate adaptive voting weights and spatial pyramid configurations to include spatial information into the geodesic histogram feature and show that this further improves results. The feature is generic and can be used as part of various algorithms. In experiments, we test the geodesic histogram feature by incorporating it into two existing video segmentation frameworks. This leads to significantly better performance in 3D video segmentation benchmarks on two datasets.
We propose a novel GAN-based framework for detecting shadows in images, in which a shadow detecti... more We propose a novel GAN-based framework for detecting shadows in images, in which a shadow detection network (D-Net) is trained together with a shadow attenuation network (A-Net) that generates adversarial training examples. The A-Net modifies the original training images constrained by a simplified physical shadow model and is focused on fooling the D-Net’s shadow predictions. Hence, it is effectively augmenting the training data for D-Net with hard-to-predict cases. The D-Net is trained to predict shadows in both original images and generated images from the A-Net. Our experimental results show that the additional training data from A-Net significantly improves the shadow detection accuracy of D-Net. Our method outperforms the stateof-the-art methods on the most challenging shadow detection benchmark (SBU) and also obtains state-of-the-art results on a cross-dataset task, testing on UCF. Furthermore, the proposed method achieves accurate real-time shadow detection at 45 frames per ...
IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021
We propose a novel deep learning method for shadow removal. Inspired by physical models of shadow... more We propose a novel deep learning method for shadow removal. Inspired by physical models of shadow formation, we use a linear illumination transformation to model the shadow effects in the image that allows the shadow image to be expressed as a combination of the shadow-free image, the shadow parameters, and a matte layer. We use two deep networks, namely SP-Net and M-Net, to predict the shadow parameters and the shadow matte respectively. This system allows us to remove the shadow effects from images. We then employ an inpainting network, I-Net, to further refine the results. We train and test our framework on the most challenging shadow removal dataset (ISTD). Our method improves the state-of-the-art in terms of mean absolute error (MAE) for the shadow area by 20%. Furthermore, this decomposition allows us to formulate a patch-based weakly-supervised shadow removal method. This model can be trained without any shadow-free images (that are cumbersome to acquire) and achieves competitive shadow removal results compared to state-of-the-art methods that are trained with fully paired shadow and shadow-free images. Last, we introduce SBU-Timelapse, a video shadow removal dataset for evaluating shadow removal methods.
We propose a novel GAN-based framework for detecting shadows in images, in which a shadow detecti... more We propose a novel GAN-based framework for detecting shadows in images, in which a shadow detection network (D-Net) is trained together with a shadow attenuation network (A-Net) that generates adversarial training examples. The A-Net modifies the original training images constrained by a simplified physical shadow model and is focused on fooling the D-Net's shadow predictions. Hence, it is effectively augmenting the training data for D-Net with hard-to-predict cases. The D-Net is trained to predict shadows in both original images and generated images from the A-Net. Our experimental results show that the additional training data from A-Net significantly improves the shadow detection accuracy of D-Net. Our method outperforms the stateof-the-art methods on the most challenging shadow detection benchmark (SBU) and also obtains state-of-the-art results on a cross-dataset task, testing on UCF. Furthermore, the proposed method achieves accurate real-time shadow detection at 45 frames per second.
Proceedings on Privacy Enhancing Technologies, 2020
In this paper, we present a large-scale measurement study of the smart TV advertising and trackin... more In this paper, we present a large-scale measurement study of the smart TV advertising and tracking ecosystem. First, we illuminate the network behavior of smart TVs as used in the wild by analyzing network traffic collected from residential gateways. We find that smart TVs connect to well-known and platform-specific advertising and tracking services (ATSes). Second, we design and implement software tools that systematically explore and collect traffic from the top-1000 apps on two popular smart TV platforms, Roku and Amazon Fire TV. We discover that a subset of apps communicate with a large number of ATSes, and that some ATS organizations only appear on certain platforms, showing a possible segmentation of the smart TV ATS ecosystem across platforms. Third, we evaluate the (in)effectiveness of DNS-based blocklists in preventing smart TVs from accessing ATSes. We highlight that even smart TV-specific blocklists suffer from missed ads and incur functionality breakage. Finally, we exam...
2017 IEEE International Conference on Computer Vision Workshops (ICCVW), 2017
Co-localization is the problem of localizing categorical objects using only positive set of examp... more Co-localization is the problem of localizing categorical objects using only positive set of example images, without any form of further supervision. This is a challenging task as there is no pixel-level annotations. Motivated by human visual learning, we find the common features of an object category from convolutional kernels of a pretrained convolutional neural network (CNN). We call these category-consistent CNN features. Then, we co-propagate their activated spatial regions using superpixel geodesic distances for localization. In our first set of experiments, we show that the proposed method achieves state-of-the-art performance on three related benchmarks: PASCAL 2007, PASCAL-2012, and the Object Discovery dataset. We also show that our method is able to detect and localize truly unseen categories, using six held-out ImagNet subset of categories with state-of-the-art accuracies. Our intuitive approach achieves this success without any region proposals or object detectors, and can be based on a CNN that was pre-trained purely on image classification tasks without further fine-tuning.
In this work, we tackle the problem of crowd counting in images. We present a Convolutional Neura... more In this work, we tackle the problem of crowd counting in images. We present a Convolutional Neural Network (CNN) based density estimation approach to solve this problem. Predicting a high resolution density map in one go is a challenging task. Hence, we present a two branch CNN architecture for generating high resolution density maps, where the first branch generates a low resolution density map, and the second branch incorporates the low resolution prediction and feature maps from the first branch to generate a high resolution density map. We also propose a multi-stage extension of our approach where each stage in the pipeline utilizes the predictions from all the previous stages. Empirical comparison with the previous state-of-the-art crowd counting methods shows that our method achieves the lowest mean absolute error on three challenging crowd counting benchmarks: Shanghaitech, World-Expo'10, and UCF datasets.
2019 IEEE/CVF International Conference on Computer Vision (ICCV), 2019
We propose a novel deep learning method for shadow removal. Inspired by physical models of shadow... more We propose a novel deep learning method for shadow removal. Inspired by physical models of shadow formation, we use a linear illumination transformation to model the shadow effects in the image that allows the shadow image to be expressed as a combination of the shadow-free image, the shadow parameters, and a matte layer. We use two deep networks, namely SP-Net and M-Net, to predict the shadow parameters and the shadow matte respectively. This system allows us to remove the shadow effects on the images. We train and test our framework on the most challenging shadow removal dataset (ISTD). Compared to the state-ofthe-art method, our model achieves a 40% error reduction in terms of root mean square error (RMSE) for the shadow area, reducing RMSE from 13.3 to 7.9. Moreover, we create an augmented ISTD dataset based on an image decomposition system by modifying the shadow parameters to generate new synthetic shadow images. Training our model on this new augmented ISTD dataset further lowers the RMSE on the shadow area to 7.4.
Nowadays, deep learning is becoming increasingly important in our daily life. The appearance of d... more Nowadays, deep learning is becoming increasingly important in our daily life. The appearance of deep learning in many applications in life relates to prediction and classification such as self-driving, product recommendation, advertisements and healthcare. Therefore, if a deep learning model causes false predictions and misclassification, it can do great harm. This is basically a crucial issue in the deep learning model. In addition, deep learning models use large amounts of data in the training/learning phases, which contain sensitive information. Therefore, when deep learning models are used in real-world applications, it is required to protect the privacy information used in the model. In this article, we carry out a brief review of the threats and defenses methods on security issues for the deep learning models and the privacy of the data used in such models while maintaining their performance and accuracy. Finally, we discuss current challenges and future developments.
Proceedings of the Fifth International Conference on Web Information Systems and Technologies, 2009
This paper studies the problem of classifying structured data sources on the Web. While prior wor... more This paper studies the problem of classifying structured data sources on the Web. While prior works use all features, once extracted from search interfaces, we further refine the feature set. In our research, each search interface is treated simply as a bag-of-words. We choose a subset of words, which is suited to classify web sources, by our feature selection methods with new metrics and a novel simple ranking scheme. Using aggressive feature selection approach, together with a Gaussian process classifier, we obtained high classification performance in an evaluation over real web data.
Uploads
Papers by Hieu Le