GANs are designed to learn a single distribution, though multiple distributions can be modeled by treating them separately. However, this naive implementation does not consider overlapping distributions. We propose Mixed Membership Generative Adversarial Networks (MMGAN) analogous to mixed-membership models that model multiple distributions and discover their commonalities and particularities. Each data distribution is modeled as a mixture over a common set of generator distributions, and mixture weights are automatically learned from the data. Mixture weights can give insight into common and unique features of each data distribution. We evaluate our proposed MMGAN and show its effectiveness on MNIST and Fashion-MNIST with various settings.

Computationally intensive simulations have made analog circuit sizing challenging for complicated analog circuit performance characterization. Accurate yet computationally efficient data-driven models of circuit performance can potentially accelerate the design and verification process. However, as analog circuits are designed under strict functional and technology constraints, there is a scarcity of data for analog circuit performance classification, posing challenges to data-driven approaches; acquiring more data typically involves running expensive and time consuming simulations. We propose Bayesian Deep Active Learning (BDAL) to learn models using fewer simulations, by iteratively selecting a small number of informative samples to label based on the model uncertainty. Bayesian neural networks used in the BDAL framework are better able to model weight uncertainty while being sufficiently expressive to model complex circuits. Compared with the state-of-the-art approaches, the proposed BDAL method can obtain better classification performance with much fewer number of simulations. Experiments on four diverse analog circuits demonstrate BDAL can achieve significant reduction in data requirement and obtain similar performance with much less labeled data for analog circuit performance classification.

In this paper, we present a coding framework for deep convolutional neural network compression. Our approach utilizes the classical coding theories and formulates the compression of deep convolutional neural networks as a rate-distortion optimization problem. We incorporate three coding ingredients in the coding framework, including bit allocation, dead zone quantization, and Tunstall coding, to improve the rate-distortion frontier without noticeable system-level overhead introduced. Experimental results show that our approach achieves state-of-the-art results on various deep convolutional neural networks and obtains considerable speedup on two deep learning accelerators. Specifically, our approach achieves 20× compression ratio on ResNet-18, ResNet-34, and ResNet-50, and 10× compression ratio on the compact already model MobileNet-v2, without hurting the accuracy. We then examine the system level impact of our approach when deploying the compressed models to hardware platforms. Hardware simulation results show that our approach obtains up to 4.3× and 2.8× inference speedup on state-of-the-art deep learning accelerators TPU and Eyeriss, respectively.

Fully homomorphic encryption (FHE) is a powerful cryptographic primitive to secure outsourced computations against an untrusted third-party provider. With the growing demand for AI and the usefulness of machine learning as a service (MLaaS), the need for secure training and inference of artificial neural networks is rising. However, the computational complexity of existing FHE schemes has been a strong deterrent to this. Prior works suffered from accuracy degradation, lack of scalability, and ciphertext expansion issues. In this paper, we take the first step towards the problem of space-efficiency in evaluating deep neural networks through designing DOReN: a low depth, batched neuron that can simultaneously evaluate multiple quantized ReLU-activated neurons on encrypted data without approximations. Our circuit design reduced the complexity of the accumulator circuit depth from O(logm ·logn) to O(logm + logn) for n bit integers. The experimental results show that the amortized processing time of our homomorphic neuron is approximately 1.26 seconds for 300 inputs and less than 0.13 seconds for 10 inputs at 80 bit security, which is a 20 fold improvement upon Lou and Jiang, NeurIPS 2019.

We examine two key questions in GAN training, namely overfitting and mode drop, from an empirical perspective. We show that when stochasticity is removed from the training procedure, GANs can overfit and exhibit almost no mode drop. Our results shed light on important characteristics of the GAN training procedure. They also provide evidence against prevailing intuitions that GANs do not memorize the training set, and that mode dropping is mainly due to properties of the GAN objective rather than how it is optimized during training.

With the increasing global popularity of self-driving cars, there is an immediate need for challenging real-world datasets for benchmarking and training various computer vision tasks such as 3D object detection. Existing datasets either represent simple scenarios or provide only day-time data. In this paper, we introduce a new challenging A*3D dataset which consists of RGB images and LiDAR data with a significant diversity of scene, time, and weather. The dataset consists of high-density images (≈ 10 times more than the pioneering KITTI dataset), heavy occlusions, a large number of nighttime frames (≈ 3 times the nuScenes dataset), addressing the gaps in the existing datasets to push the boundaries of tasks in autonomous driving research to more challenging highly diverse environments. The dataset contains 39K frames, 7 classes, and 230K 3D object annotations. An extensive 3D object detection benchmark evaluation on the A*3D dataset for various attributes such as high density, day-time/night-time, gives interesting insights into the advantages and limitations of training and testing 3D object detection in real-world setting.

Answering questions using multi-modal context is a challenging problem as it requires a deep integration of diverse data sources. Existing approaches only consider a subset of all possible interactions among data sources during one attention hop. In this paper, we present a Holistic Multi-modal Memory Network (HMMN) framework that fully considers interactions between different input sources (multi-modal context, question) at each hop. In addition, to hone in on relevant information, our framework takes answer choices into consideration during the context retrieval stage. Our HMMN framework effectively integrates information from the multi-modal context, question, and answer choices, enabling more informative context to be retrieved for question answering. Experimental results on the MovieQA and TVQA datasets validate the effectiveness of our HMMN framework. Extensive ablation studies show the importance of holistic reasoning and reveal the contributions of different attention strategies to model performance.

Modern convolutional object detectors have improved the detection accuracy significantly, which in turn inspired the development of dedicated hardware accelerators to achieve real-time performance by exploiting inherent parallelism in the algorithm. Non-maximum suppression (NMS) is an indispensable operation in object detection. In stark contrast to most operations, the commonly-adopted GreedyNMS algorithm does not foster parallelism, which can be a major performance bottleneck. In this paper, we introduce MaxpoolNMS, a parallelizable alternative to the NMS algorithm, which is based on max-pooling classification score maps. By employing a novel multi-scale multi-channel max-pooling strategy, our method is 20x faster than GreedyNMS while simultaneously achieves comparable accuracy, when quantified across various benchmarking datasets, i.e., MS COCO, KITTI and PASCAL VOC. Furthermore, our method is better suited for hardware-based acceleration than GreedyNMS.

This paper addresses a challenging problem - how to reduce energy consumption without incurring performance drop when deploying deep neural networks (DNNs) at the inference stage. In order to alleviate the computation and storage burdens, we propose a novel dataflow-based joint quantization approach with the hypothesis that a fewer number of quantization operations would incur less information loss and thus improve the final performance. It first introduces a quantization scheme with efficient bit-shifting and rounding operations to represent network parameters and activations in low precision. Then it re-structures the network architectures to form unified modules for optimization on the quantized model. Extensive experiments on ImageNet and KITTI validate the effectiveness of our model, demonstrating that state-of-the-art results for various tasks can be achieved by this quantized model. Besides, we designed and synthesized an RTL model to measure the hardware costs among various quantization methods. For each quantization operation, it reduces area cost by about 15 times and energy consumption by about 9 times, compared to a strong baseline.

In this paper, we propose Deep Holographic Networks (DHN) to learn similarity metrics of videos for multi-label video search. DHN introduces a holographic composition layer to explicitly encode similarity metrics at intermediate layer of the network, instead of conventional deep metric learning approaches driven by ranking losses. The holographic composition layer is parameter-free and enables less memory footprint compared with state-of-the-art. Towards multi-label video search at large scale, we present a new video benchmark built upon the YouTube-8M dataset. Extensive evaluations on this dataset demonstrate that DHN performs better than traditional deep metric learning approaches as well as other compositional networks.

For most real-time implementations of deep artificial neural networks, both weights and intermediate layer activations have to be stored and loaded for processing. Compression of both is often advisable to mitigate the memory bottleneck. In this paper, we propose a bit allocation framework for compressing the weights and activations of deep neural networks. The differentiability of input-output relationships for all network layers allows us to relate the neural network output accuracy to the bit-rate of quantized weights and layer activations. We formulate a Lagrangian optimization framework that finds the optimum joint bit allocation among all intermediate activation layers and weights. Our method obtains excellent results on two deep neural networks, VGG-16 and ResNet-50. Without requiring re-training, it outperforms other state-of-the-art neural network compression methods.

Embedded deep learning platforms have witnessed two simultaneous improvements. First, the accuracy of convolutional neural networks (CNNs) has been significantly improved through the use of automated neural-architecture search (NAS) algorithms to determine CNN structure. Second, there has been increasing interest in developing hardware accelerators for CNNs that provide improved inference performance and energy consumption compared to GPUs. Such embedded deep learning platforms differ in the amount of compute resources and memory-access bandwidth, which would affect performance and energy consumption of CNNs. It is therefore critical to consider the available hardware resources in the network architecture search. To this end, we introduce TEA-DNN, a NAS algorithm targeting multi-objective optimization of execution time, energy consumption, and classification accuracy of CNN workloads on embedded architectures. TEA-DNN leverages energy and execution time measurements on embedded hardware when exploring the Pareto-optimal curves across accuracy, execution time, and energy consumption and does not require additional effort to model the underlying hardware. We apply TEA-DNN for image classification on actual embedded platforms (NVIDIA Jetson TX2 and Intel Movidius Neural Compute Stick). We highlight the Pareto-optimal operating points that emphasize the necessity to explicitly consider hardware characteristics in the search process. To the best of our knowledge, this is the most comprehensive study of Pareto-optimal models across a range of hardware platforms using actual measurements on hardware to obtain objective values.

The MPEG compact descriptors for visual search (CDVS) is a standard toward image matching and retrieval. To achieve high retrieval accuracy over a large scale image/video dataset, recent research efforts have demonstrated that employing extremely high-dimensional descriptors such as the Fisher vector (FV) and the vector of locally aggregated descriptors (VLAD) can yield good performance. Since the FV (or VLAD) possesses high discriminability but small visual vocabulary, it has been adopted by CDVS to construct a global compact descriptor. In this paper, we study the development of global compact descriptors in the completed CDVS standard and the emerging compact descriptors for video analysis (CDVA) standard, in which we formulate the FV (or VLAD) compression as a resource-constrained optimization problem. Accordingly, we propose a codebook-free aggregation method via dual selection to generate a global compact visual descriptor, which supports fast and accurate feature matching free of large visual codebooks, fulfilling the low memory requirement of mobile visual search at significantly reduced latency. Specifically, we investigate both sample-specific Gaussian component redundancy and bit dependency within a binary aggregated descriptor to produce compact binary codes. Our technique contributes to the scalable compressed Fisher vector (SCFV) adopted by the CDVS standard. Moreover, the SCFV descriptor is currently serving as the frame-level hand-crafted video feature, which inspires the inheritance of CDVS descriptors for the emerging CDVA standard. Furthermore, we investigate the positive complementary effect of our standard compliant compact descriptor and deep learning based features extracted from convolutional neural networks with significant mean average precision gains. Extensive evaluation over benchmark databases shows the significant merits of the codebook-free binary codes for scalable visual search.

Anomaly detection is a significant and hence well-studied problem. However, developing effective anomaly detection methods for complex and high-dimensional data remains a challenge. As Generative Adversarial Networks (GANs) are able to model the complex high-dimensional distributions of real-world data, they offer a promising approach to address this challenge. In this work, we propose an anomaly detection method, Adversarially Learned Anomaly Detection (ALAD) based on bi-directional GANs, that derives adversarially learned features for the anomaly detection task. ALAD then uses reconstruction errors based on these adversarially learned features to determine if a data sample is anomalous. ALAD builds on recent advances to ensure data-space and latent-space cycle-consistencies and stabilize GAN training, which results in significantly improved anomaly detection performance. ALAD achieves state-of-the-art performance on a range of image and tabular datasets while being several hundred-fold faster at test time than the only published GAN-based method.

Recently Convolutional Neural Networks (CNNs) have achieved great success in different fields including image instance retrieval. However traditional global pooling approaches fail to capture all possible discriminative information of CNN activations and treat activations over channels equally regardless of the different importance between channels. In this work, we focus on the mentioned problem of global feature pooling over CNN activations for image instance retrieval. We make two contributions. First, we introduce a channel-wise SQUare-root (SQU) pooling (2-norm) approach, which makes better use of information over activation maps and is superior to Average (1-norm) and Max pooling (infinity norm), in the context of instance retrieval. Second, we further improve SQU by learning a gating function that weights the contributions of different channels, in an end-to-end manner. Extensive experiments on 6 benchmark datasets show that the proposed strategies achieve considerable improvements over state-of-the-art.

Single image dehazing has been a classic topic in computer vision for years. Motivated by the atmospheric scattering model, the key to satisfactory single image dehazing relies on an estimation of two physical parameters, i.e., the global atmospheric light and the transmission coefficient. Most existing methods employ a two-step pipeline to estimate these two parameters with heuristics which accumulate errors and compromise dehazing quality. Inspired by differentiable programming, we re-formulate the atmospheric scattering model into a novel generative adversarial network (DehazeGAN). Such a reformulation and adversarial learning allow the two parameters to be learned simultaneously and automatically from data by optimizing the final dehazing performance so that clean images with faithful color and structures are directly produced. Moreover, our reformulation also greatly improves the GAN’s interpretability and quality for single image dehazing. To the best of our knowledge, our method is one of the first works to explore the connection among generative adversarial models, image dehazing, and differentiable programming, which advance the theories and application of these areas. Extensive experiments on synthetic and realistic data show that our method outperforms state-of-the-art methods in terms of PSNR, SSIM, and subjective visual quality.

In past years, deep convolutional neural networks (DCNN) have achieved big successes in image classification and object detection, as demonstrated on ImageNet in academic field. However, There are some unique practical challenges remain for real-world image recognition applications, e.g., small size of the objects, imbalanced data distributions, limited labeled data samples, etc. In this work, we are making efforts to deal with these challenges through a computational framework by incorporating latest developments in deep learning. In terms of two-stage detection scheme, pseudo labeling, data augmentation, cross-validation and ensemble learning, the proposed framework aims to achieve better performances for practical image recognition applications as compared to using standard deep learning methods. The proposed framework has recently been deployed as the key kernel for several image recognition competitions organized by Kaggle. The performance is promising as our final private scores were ranked 4 out of 2293 teams for fish recognition on the challenge "The Nature Conservancy Fisheries Monitoring" and 3 out of 834 teams for cervix recognition on the challenge "Intel &MobileODT Cervical Cancer Screening", and several others. We believe that by sharing the solutions, we can further promote the applications of deep learning techniques.

Deep neural networks have recently achieved competitive accuracy for human activity recognition. However, there is room for improvement, especially in modeling of long-term temporal importance and determining the activity relevance of different temporal segments in a video. To address this problem, we propose a learnable and differentiable module: Deep Adaptive Temporal Pooling (DATP). DATP applies a self-attention mechanism to adaptively pool the classification scores of different video segments. Specifically, using frame-level features, DATP regresses importance of different temporal segments, and generates weights for them. Remarkably, DATP is trained using only the video-level label. There is no need of additional supervision except video-level activity class label. We conduct extensive experiments to investigate various input features and different weight models. Experimental results show that DATP can learn to assign large weights to key video segments. More importantly, DATP can improve training of frame-level feature extractor. This is because relevant temporal segments are assigned large weights during back-propagation. Overall, we achieve state-of-the-art performance on UCF101, HMDB51 and Kinetics datasets.

Image instance retrieval is the problem of retrieving images from a database which contain the same object. Convolutional Neural Network (CNN) based descriptors are becoming the dominant approach for generating global image descriptors for the instance retrieval problem. One major drawback of CNN-based global descriptors is that uncompressed deep neural network models require hundreds of megabytes of storage making them inconvenient to deploy in mobile applications or in custom hardware. In this work, we study the problem of neural network model compression focusing on the image instance retrieval task. We study quantization, coding, pruning and weight sharing techniques for reducing model size for the instance retrieval problem. We provide extensive experimental results on the trade-off between retrieval performance and model size for different types of networks on several data sets providing the most comprehensive study on this topic. We compress models to the order of a few MBs: two orders of magnitude smaller than the uncompressed models while achieving negligible loss in retrieval performance1.

With emerging demand for large-scale video analysis, the Motion Picture Experts Group (MPEG) initiated the Compact Descriptor for Video Analysis (CDVA) standardization in 2014. In this work, we develop novel deep-learning features and incorporate them into the well-established CDVA evaluation framework to study its effectiveness in video analysis. In particular, we propose a Nested Invariance Pooling (NIP) method to obtain compact and robust Convolutional Neural Network (CNNs) descriptors. The CNNs descriptors are generated by applying three different pooling operations to the feature maps of CNNs in a nested way towards rotation and scale invariant feature representation. In particular, the rational, advantages and performance on the combination of CNNs and handcrafted descriptors are provided to better investigate the complementary effects of deep learnt and handcrafted features. Extensive experimental results show that the proposed CNNs descriptors outperform both state-of-the-art CNNs descriptors and canonical handcrafted descriptors adopted in CDVA Experimental Model (CXM) with significant mAP gains of 11.3% and 4.7%, respectively. Moreover, the combination of NIP derived deep invariant descriptors and handcrafted descriptors not only fulfills the lowest bitrate budget of CDVA, but also significantly advances the performance of CDVA core techniques.

In this work, we study the problem of deep global descriptors for video matching with regional feature pooling. We aim to analyze the joint effect of ROI (Region of Interest) size and pooling moment on video matching performance. To this end, we propose to mathematically model the distribution of video matching function with a pooling function nested in. Matching performance can be estimated by the separability of these class-conditional distributions between matching and non-matching pairs. Empirical studies on the challenging MPEG CDVA dataset demonstrate that performance trends are consistent with the estimation and experimental results, though the theoretical model is largely simplified compared to video matching and retrieval in practice.

This paper proposes a novel algorithm to solve the exact r-neighbour search problem in Hamming space. Existing r-neighbour search methods typically adopt hash table to index binary codes. Given a query, existing approaches search the nearest neighbours by checking all buckets of a Hamming ball centered at the query. The problem is these methods spend most of search time visiting empty buckets (lookup misses). In this paper, we adopt trie structure to index binary codes. We consider several continuous bits of a binary string as a block and use it as an atomic indexing element in trie structure, which is efficient in access speed and memory usage. Our method searches the nearest neighbours of a query by utilizing the records of nodes in trie structure to avoid lookup misses. We name the proposed indexing structure as Multi-Block N-ary Trie (MBNT). A theoretical analysis is given to prove that MBNT has less time cost than other hash-based methods. Extensive results show that MBNT outperforms state-of-the-art algorithms on several large scale benchmarks.

Object detection has been a key task in computer vision with deep convolutional neural networks being a significant performer. We propose a method named Region Average Pooling that leverages object co-occurrence to improve object detection performance. Given regions of interest in an image, our method augments object detection networks with pooled contextual features from other regions of interest in the scene. We implement our scheme and evaluate it on the Pascal Visual Object Classes (VOC) 2007 and Microsoft Common Objects in Context (MS COCO) datasets. When used as part of the Faster R-CNN object detection framework with VGG-16, we show an increase in mAP from 24.2% to 25.5% over baseline Faster R-CNN and Global Average Pooling when testing on MS COCO.

Object detection in images is a crucial task in computer vision, with important applications ranging from security surveillance to autonomous vehicles. Existing state-of-the-art algorithms, including deep neural networks, only focus on utilizing features within an image itself, largely neglecting the vast amount of background knowledge about the real world. In this paper, we propose a novel framework of knowledge-aware object detection, which enables the integration of external knowledge such as knowledge graphs into any object detection algorithm. The framework employs the notion of semantic consistency to quantify and generalize knowledge, which improves object detection through a re-optimization process to achieve better consistency with background knowledge. Finally, empirical evaluation on two benchmark datasets show that our approach can significantly increase recall by up to 6.3 points without compromising mean average precision, when compared to the state-of-the-art baseline.

This work focuses on representing very high-dimensional global image descriptors using very compact 64-1024 bit binary hashes for instance retrieval. We propose DeepHash: a hashing scheme based on deep networks. Key to making DeepHash work at extremely low bitrates are three important considerations -- regularization, depth and fine-tuning -- each requiring solutions specific to the hashing problem. In-depth evaluation shows that our scheme outperforms state-of-the-art methods over several benchmark datasets for both Fisher Vectors and Deep Convolutional Neural Network features, by up to 8.5% over other schemes. The retrieval performance with 256-bit hashes is close to that of the uncompressed floating point features -- a remarkable 512x compression.

The goal of this work is the computation of very compact binary hashes for image instance retrieval. Our approach has two novel contributions. The first one is Nested Invariance Pooling (NIP), a method inspired from i-theory, a mathematical theory for computing group invariant transformations with feed-forward neural networks. NIP is able to produce compact and well-performing descriptors with visual representations extracted from convolutional neural networks. We specifically incorporate scale, translation and rotation invariances but the scheme can be extended to any arbitrary sets of transformations. We also show that using moments of increasing order throughout nesting is important. The NIP descriptors are then hashed to the target code size (32-256 bits) with a Restricted Boltzmann Machine with a novel batch-level regularization scheme specifically designed for the purpose of hashing (RBMH). A thorough empirical evaluation with state-of-the-art shows that the results obtained both with the NIP descriptors and the NIP+RBMH hashes are consistently outstanding across a wide range of datasets.

Inspired by progresses in cognitive science, artificial intelligence, computer vision, and mobile computing technologies, we propose and implement a wearable virtual usher for cognitive indoor navigation based on egocentric visual perception. A novel computational framework of cognitive wayfinding in an indoor environment is proposed, which contains a context model, a route model, and a process model. A hierarchical structure is proposed to represent the cognitive context knowledge of indoor scenes. Given a start position and a destination, a Bayesian network model is proposed to represent the navigation route derived from the context model. A novel dynamic Bayesian network (DBN) model is proposed to accommodate the dynamic process of navigation based on real-time first-person-view visual input, which involves multiple asynchronous temporal dependencies. To adapt to large variations in travel time through trip segments, we propose an online adaptation algorithm for the DBN model, leading to a self-adaptive DBN. A prototype system is built and tested for technical performance and user experience. The quantitative evaluation shows that our method achieves over 13% improvement in accuracy as compared to baseline approaches based on hidden Markov model. In the user study, our system guides the participants to their destinations, emulating a human usher in multiple aspects.

With emerging demand for large-scale video analysis, MPEG initiated the compact descriptor for video analysis (CDVA) standardization in 2014. Beyond handcrafted descriptors adopted by the current MPEG-CDVA reference model, we study the problem of deep learned global descriptors for video matching, localization, and retrieval. First, inspired by a recent invariance theory, we propose a nested invariance pooling (NIP) method to derive compact deep global descriptors from convolutional neural networks (CNNs), by progressively encoding translation, scale, and rotation invariances into the pooled descriptors. Second, our empirical studies have shown that a sequence of well designed pooling moments (e.g., max or average) may drastically impact video matching performance, which motivates us to design hybrid pooling operations via NIP (HNIP). HNIP has further improved the discriminability of deep global descriptors. Third, the technical merits and performance improvements by combining deep and handcrafted descriptors are provided to better investigate the complementary effects. We evaluate the effectiveness of HNIP within the well-established MPEG-CDVA evaluation framework. The extensive experiments have demonstrated that HNIP outperforms the state-of-the-art deep and canonical handcrafted descriptors with significant mAP gains of 5.5% and 4.7%, respectively. In particular the combination of HNIP incorporated and handcrafted global descriptors has significantly boosted the performance of CDVA core techniques with comparable descriptor size.

A typical image retrieval pipeline starts with the comparison of global descriptors from a large database to find a short list of candidate matches. A good image descriptor is key to the retrieval pipeline and should reconcile two contradictory requirements: providing recall rates as high as possible and being as compact as possible for fast matching. Following the recent successes of Deep Convolutional Neural Networks (DCNN) for large scale image classification, descriptors extracted from DCNNs are increasingly used in place of the traditional hand crafted descriptors such as Fisher Vectors (FV) with better retrieval performances. Nevertheless, the dimensionality of a typical DCNN descriptor-extracted either from the visual feature pyramid or the fully-connected layers-remains quite high at several thousands of scalar values. In this paper, we propose Unsupervised Triplet Hashing (UTH), a fully unsupervised method to compute extremely compact binary hashes-in the 32-256 bits range-from high-dimensional global descriptors. UTH consists of two successive deep learning steps. First, Stacked Restricted Boltzmann Machines (SRBM), a type of unsupervised deep neural nets, are used to learn binary embedding functions able to bring the descriptor size down to the desired bitrate. SRBMs are typically able to ensure a very high compression rate at the expense of loosing some desirable metric properties of the original DCNN descriptor space. Then, triplet networks, a rank learning scheme based on weight sharing nets is used to fine-tune the binary embedding functions to retain as much as possible of the useful metric properties of the original space. A thorough empirical evaluation conducted on multiple publicly available dataset using DCNN descriptors shows that our method is able to significantly outperform state-of-the-art unsupervised schemes in the target bit range.

With the increasing availability of wearable devices, research on egocentric activity recognition has received much attention recently. In this paper, we build a Multimodal Egocentric Activity dataset which includes egocentric videos and sensor data of 20 fine-grained and diverse activity categories. We present a novel strategy to extract temporal trajectory-like features from sensor data. We propose to apply the Fisher Kernel framework to fuse video and temporal enhanced sensor features. Experiment results show that with careful design of feature extraction and fusion algorithm, sensor data can enhance information-rich video data. We make publicly available the Multimodal Egocentric Activity dataset to facilitate future research.

Compact descriptors for visual search (CDVS) is a recently completed standard from the ISO/IEC moving pictures experts group (MPEG). The primary goal of this standard is to provide a standardized bitstream syntax to enable interoperability in the context of image retrieval applications. Over the course of the standardization process, remarkable improvements were achieved in reducing the size of image feature data and in reducing the computation and memory footprint in the feature extraction process. This paper provides an overview of the technical features of the MPEG-CDVS standard and summarizes its evolution.

The first step in an image retrieval pipeline consists of comparing global descriptors from a large database to find a short list of candidate matching images. The more compact the global descriptor, the faster the descriptors can be compared for matching. State-of-the-art global descriptors based on Fisher Vectors are represented with tens of thousands of floating point numbers. While there is significant work on compression of local descriptors, there is relatively little work on compression of high dimensional Fisher Vectors. We study the problem of global descriptor compression in the context of image retrieval, focusing on extremely compact binary representations: 64-1024 bits. Motivated by the remarkable success of deep neural networks in recent literature, we propose a compression scheme based on deeply stacked Restricted Boltzmann Machines (SRBM), which learn lower dimensional non-linear subspaces on which the data lie. We provide a thorough evaluation of several state-of-the-art compression schemes based on PCA, Locality Sensitive Hashing, Product Quantization and greedy bit selection, and show that the proposed compression scheme outperforms all existing schemes.

Results of the ICDAR 2015 Robust Reading Competition are presented. A new Challenge 4 on Incidental Scene Text has been added to the Challenges on Born-Digital Images, Focused Scene Images and Video Text. Challenge 4 is run on a newly acquired dataset of 1,670 images evaluating Text Localisation, Word Recognition and End-to-End pipelines. In addition, the dataset for Challenge 3 on Video Text has been substantially updated with more video sequences and more accurate ground truth data. Finally, tasks assessing End-to-End system performance have been introduced to all Challenges. The competition took place in the first quarter of 2015, and received a total of 44 submissions. Only the tasks newly introduced in 2015 are reported on. The datasets, the ground truth specification and the evaluation protocols are presented together with the results and a brief summary of the participating methods.

In this work, we propose to divide each class (a person) into subclasses using spatial partition trees which helps in better capturing the intra-personal variances arising from the appearances of the same individual. We perform a comprehensive analysis on within-class and within-subclass eigen-spectrums of face images and propose a novel method of eigen-spectrum modeling which extracts discriminative features of faces from both within-subclass and total or between-subclass scatter matrices. Effective low-dimensional face discriminative features are extracted for face recognition (FR) after performing discriminant evaluation in the entire eigenspace. Experimental results on popular face databases (AR, FERET) and the challenging unconstrained YouTube Face database show the superiority of our proposed approach on all three databases.

Compact keyframe-based video summaries are a popular way of generating viewership on video sharing platforms. Yet, creating relevant and compelling summaries for arbitrarily long videos with a small number of keyframes is a challenging task. We propose a comprehensive keyframe-based summarization framework combining deep convolutional neural networks and restricted Boltzmann machines. An original co-regularization scheme is used to discover meaningful subject-scene associations. The resulting multimodal representations are then used to select highly-relevant keyframes. A comprehensive user study is conducted comparing our proposed method to a variety of schemes, including the summarization currently in use by one of the most popular video sharing websites. The results show that our method consistently outperforms the baseline schemes for any given amount of keyframes both in terms of attractiveness and in-formativeness. The lead is even more significant for smaller summaries.

Presentations have been an effective means of delivering information to groups for ages. Over the past few decades, technological advancements have revolutionized the way humans deliver presentations. Despite that, the quality of presentations can be varied and affected by a variety of reasons. Conventional presentation evaluation usually requires painstaking manual analysis by experts. Although the expert feedback can definitely assist users in improving their presentation skills, manual evaluation suffers from high cost and is often not accessible to most people. In this work, we propose a novel multi-sensor self-quantification framework for presentations. Utilizing conventional ambient sensors (i.e., static cameras, Kinect sensor) and the emerging wearable egocentric sensors (i.e., Google Glass), we first analyze the efficacy of each type of sensor with various nonverbal assessment rubrics, which is followed by our proposed multi-sensor presentation analytics framework. The proposed framework is evaluated on a new presentation dataset, namely NUS Multi-Sensor Presentation (NUSMSP) dataset, which consists of 51 presentations covering a diverse set of topics. The dataset was recorded with ambient static cameras, Kinect sensor, and Google Glass. In addition to multi-sensor analytics, we have conducted a user study with the speakers to verify the effectiveness of our system generated analytics, which has received positive and promising feedback.

MPEG is currently developing a standard titled Compact Descriptors for Visual Search (CDVS) for descriptor extraction and compression. In this work, we report comprehensive patch-level experiments for a direct comparison of low bitrate descriptors for visual search. For evaluating different compression schemes, we propose a dataset of matching pairs of image patches from the MPEG-CDVS image-level data sets. We propose a greedy rate allocation scheme for distributing bits across different spatialbins of the SIFT descriptor. We study a scheme based on Entropy Constrained Vector Quantization and greedy rate allocation, which performs close to the performancebound for any compression scheme. Finally, we present extensive feature-level Receiver Operating Characteristic (ROC) comparisons for different compression schemes (VectorQuantization, Transform Coding, Lattice Coding) proposed during the MPEG-CDVS standardization process.

We study the challenges of image-based retrieval when the database consists of videos. This variation of visual search is important for a broad range of applications that require indexing video databases based on their visual contents. We present new solutions to reduce storage requirements, while at the same time improving video search quality. The video database is preprocessed to find different appearances of the same visual elements, and build robust descriptors. Compression algorithms are developed to reduce system's storage requirements. We introduce a dataset of CNN broadcasts and queries that include photos taken with mobile phones and images of objects. Our experiments include pairwise matching and retrieval scenarios. We demonstrate one order of magnitude storage reduction and search quality improvements of up to 12% in mean average precision, compared to a baseline system that does not make use of our techniques.

With wearable devices like Google Glass, it will soon become possible to record everything we see. We envision a system where one's entire visual memory is captured, stored and indexed. One of the biggest challenges is the scale of the retrieval problem. In this work, we focus on how to organize streaming egocentric video data. Egocentric video data is highly redundant, in that, we see several objects and scenes repeatedly as we go about our lives. To exploit this redundancy, we propose an evolving sparse-graph representation for egocentric video data. We propose an incremental local density clustering scheme, which learns salient objects and scenes for streaming egocentric video data. We use the density clustering scheme to prune redundant data in the database. For image-retrieval applications, by retaining only representative nodes from dense sub graphs in the streaming data source, we show we can achieve 90% of peak recall by retaining only 1% of data, with a significant 18% improvement in absolute recall over naive uniform sub sampling of the egocentric video data.

Streaming mobile augmented reality applications require both real-time recognition and tracking of objects of interest in a video sequence. Typically, local features are calculated from the gradients of a canonical patch around a keypoint in individual video frames. In this paper, we propose a temporally coherent keypoint detector and design efficient interframe predictive coding techniques for canonical patches, feature descriptors, and keypoint locations. In the proposed system, we strive to transmit each patch or its equivalent feature descriptor with as few bits as possible by modifying a previously transmitted patch or descriptor. Our solution enables server-based mobile augmented reality where a continuous stream of salient information, sufficient for image-based retrieval, and object localization, is sent at a bit-rate that is practical for today's wireless links and less than one-tenth of the bit-rate needed to stream the compressed video to the server.

We present the radial gradient transform (RGT) and a fast approximation, the approximate RGT (ARGT). We analyze the effects of the approximation on gradient quantization and histogramming. The ARGT is incorporated into the rotation-invariant fast feature (RIFF) algorithm. We demonstrate that, using the ARGT, RIFF extracts features 16× faster than SURF while achieving a similar performance for image matching and retrieval.

Local features are widely used for content-based image retrieval and object recognition. Most feature descriptors are calculated from the gradients of a canonical patch around repeatable keypoints in the image. In this paper, we propose a technique for designing quantization matrices that reduce the mean squared error distortion of the gradient derived from DCT-encoded canonical patches. Experimental results demonstrate that our proposed patch encoder greatly outperforms a JPEG encoder at the same encoding complexity. Moreover, our quantization matrices achieve lower gradient distortion and larger number of feature matches at the same bit-rate.

Local features are widely used for content-based image retrieval and augmented reality applications. Typically, feature descriptors are calculated from the gradients of a canonical patch around a repeatable key point in the image. In previous work, we showed that one can alternatively transmit the compressed canonical patch and perform descriptor computation at the receiving end with comparable performance. In this paper, we propose a temporally coherent key point detector in order to allow efficient interframe coding of canonical patches. In inter-patch compression, one strives to transmit each patch with as few bits as possible by simply modifying a previously transmitted patch. This enables server-based mobile augmented reality where a continuous stream of salient information, sufficient for the image-based retrieval, can be sent over a wireless link at the smallest possible bit-rate. Experimental results show that our technique achieves a similar image matching performance at 1/10 of the bit-rate when compared to detecting key points independently frame-by-frame.

Automatic Language Identification (LID) in music has received significantly less attention than LID in speech. Here, we study the problem of LID in music videos uploaded on YouTube. We use a “bag-of-words” approach based on state-of-the-art content based audio-visual features and linear SVM classifiers for automatic LID. Our system obtains 48% accuracy for a corpus of 25000 music videos and 25 different languages.

We survey popular data sets used in computer vision literature and point out their limitations for mobile visual search applications. To overcome many of the limitations, we propose the Stanford Mobile Visual Search data set. The data set contains camera-phone images of products, CDs, books, outdoor landmarks, business cards, text documents, museum paintings and video clips. The data set has several key characteristics lacking in existing data sets: rigid objects, widely varying lighting conditions, perspective distortion, foreground and background clutter, realistic ground-truth reference data, and query data collected from heterogeneous low and high-end camera phones. We hope that the data set will help push research forward in the field of mobile visual search.

We present a method that unifies tracking and video content recognition with applications to Mobile Augmented Reality (MAR). We introduce the Radial Gradient Transform (RGT) and an approximate RGT, yielding the Rotation-Invariant, Fast Feature (RIFF) descriptor. We demonstrate that RIFF is fast enough for real-time tracking, while robust enough for large scale retrieval tasks. At 26× the speed, our tracking-scheme obtains a more accurate global affine motion-model than the Kanade Lucas Tomasi (KLT) tracker. The same descriptors can achieve 94% retrieval accuracy from a database of 104images.

In this paper, they address a key challenge for scaling image search up to larger databases: the amount of memory consumed by the inverted index. In a VT-based image retrieval system, the most memory-intensive structure is the inverted index. For example, in a database of one million images where each image contains hundreds of features, the inverted index consumes 2.5 GB of RAM. Such large memory usage limits the ability to run other concurrent processes on the same server, such as recognition systems for other databases. A memory-congested server can exhibit swapping between main and virtual memory, which significantly slows down all processes.

We evaluate the performance of MPEG-7 image signatures, Compressed Histogram of Gradients descriptor (CHoG) and Scale Invariant Feature Transform (SIFT) descriptors for mobile visual search applications. We observe that SIFT and CHoG outperform MPEG-7 image signatures greatly in terms of feature-level Receiver Operating Characteristic (ROC) performance and image-level matching. Moreover, CHoG descriptors demonstrate such gains while being comparable with MPEG-7 image signatures in bit-rate.

In this paper, we focus on a new application of mobile visual search: snapping a photo with a mobile device of a video playing on a TV screen to automatically retrieve and stream the remainder of the video to the mobile device. When the user takes a photo of the video, the captured query frame may contain too few useful features for good retrieval performance. We design and implement a new algorithm for mobile video retrieval to accurately select a feature-rich frame from a sequence of viewfinder frames in a very short temporal window determined by the user-initiated query event. Fast and accurate selection using efficiently computed Hessian scores is developed for real-time operation on mobile devices. Viewfinder frames captured before the query starts are pre-processed, while the number of viewfinder frames captured afterwards is minimized by a probabilistic optimization process. Evaluated on a large video database of 10 million frames, dynamic query frame selection provides a substantial increase in retrieval accuracy with very low search latency.

We present a fast and efficient geometric re-ranking method that can be incorporated in a feature based image-based retrieval system that utilizes a Vocabulary Tree (VT). We form feature pairs by comparing descriptor classification paths in the VT and calculate geometric similarity score of these pairs. We propose a location geometric similarity scoring method that is invariant to rotation, scale, and translation, and can be easily incorporated in mobile visual search and augmented reality systems. We compare the performance of the location geometric scoring scheme to orientation and scale geometric scoring schemes. We show in our experiments that re-ranking schemes can substantially improve recognition accuracy. We can also reduce the worst case server latency up to 1 sec and still improve the recognition performance.

Restoring the fine detail in the slide area of a defocused lecture video is a challenging task. In this work, we propose to use clean images of slides available along with the defocused lecture video to help the restoration. Our proposed method uses local feature descriptors and multiple defocused slide decks to automatically identify the slide that is displayed in the defocused frame. We then use the matching slide as side information to estimate the parameters for deconvolution and bilateral filtering. Experimental results show that the proposed algorithm compares favorably to a computationally-intensive iterative deconvolution algorithm that does not employ any side information. In particular, it can recover small drawings and text that are severely blurred in a poorly focused lecture video

We present a mobile product recognition system for the camera-phone. By snapping a picture of a product with a camera-phone, the user can retrieve online information of the product. The product is recognized by an image-based retrieval system located on a remote server. Our database currently comprises more than one million entries, primarily products packaged in rigid boxes with printed labels, such as CDs, DVDs, and books. We extract low bit-rate descriptors from the query image and compress the location of the descriptors using location histogram coding on the camera-phone. We transmit the compressed query features, instead of a query image, to reduce the transmission delay. We use inverted index compression and fast geometric re-ranking on our database to provide a low delay image recognition response for large scale databases. Experimental timing results on different parts of the mobile product recognition system is reported in this work.

Establishing visual correspondences is an essential component of many computer vision problems, and is often done with robust, local feature-descriptors. Transmission and storage of these descriptors are of critical importance in the context of mobile distributed camera networks and large indexing problems. We propose a framework for computing low bit-rate feature descriptors with a 20× reduction in bit rate. The framework is low complexity and has significant speed-up in the matching stage. We represent gradient histograms as tree structures which can be efficiently compressed. We show how to efficiently compute distances between descriptors in their compressed representation eliminating the need for decoding. We perform a comprehensive performance comparison with SIFT, SURF, and other low bit-rate descriptors and show that our proposed CHoG descriptor outperforms existing schemes.

For mobile image matching applications, a mobile device captures a query image, extracts descriptive features, and transmits these features wirelessly to a server. The server recognizes the query image by comparing the extracted features to its database and returns information associated with the recognition result. For slow links, query feature compression is crucial for low-latency retrieval. Previous image retrieval systems transmit compressed feature descriptors, which is well suited for pairwise image matching. For fast retrieval from large databases, however, scalable vocabulary trees are commonly employed. In this paper, we propose a rate-efficient codec designed for tree-based retrieval. By encoding a tree histogram, our codec can achieve a more than 5times rate reduction compared to sending compressed feature descriptors. By discarding the order amongst a list of features, histogram coding requires 1.5times lower rate than sending a tree node index for every feature. A statistical analysis is performed to study how the entropy of encoded symbols varies with tree depth and the number of features.

Illumination distortion due to uncontrolled lighting can severely degrade the color appearance of a photo. Frequently, the desired colors for objects in a newly taken query image are found in a previously stored database image. Then, the goal is to change the colors in the query image to match the colors in the database image. This paper presents a color restoration system that automatically retrieves a database image which matches the query image, even if the two images are taken from different viewpoints and under different illuminations. Robust features enable both accurate retrieval from the database and efficient sampling of the color differences between the query and database images. A spatially varying color mismatch model is generated, and the colors of the query image are effectively restored.

We propose a novel, low-complexity, tracking scheme that uses motion vectors directly from a video coder. We compare our tracking algorithm against ground truth data, and show that we can achieve a high level of accuracy, even though the motion vectors are rate-distortion optimized and do not represent true motion. We develop a framework for tracking in video sequences with various GOP structures. Such a scheme would find applications in the context of Mobile Augmented Reality. The proposed feature tracking algorithm can significantly reduce the required rate of feature extraction and matching.

For large wireless sensor networks, identifying the exact location of every sensor may not be feasible or necessary. A coarse estimate of the sensors' locations is usually sufficient for many applications. An efficient area localization scheme (ALS) has been proposed for large sensor networks. ALS is a range-free localization scheme that tries to estimate the position of a sensor within a certain area rather than its exact location. As the complex calculations are handled by the powerful sinks instead of the sensors, this reduces the energy consumed by the sensors and helps to extend the lifetime of the network. In this paper, we discuss the implementation of ALS using MICAz motes and the localization experiments carried out in both indoor and outdoor environments. We also propose algorithms and techniques to improve the accuracy of ALS in realistic deployment scenarios. We observed that the performance of ALS is comparable or better than other localization schemes that have been implemented while ALS has lower complexity and hardware requirements.

Iterative multilateration techniques have been proposed to improve position estimates in localization schemes for sensor networks. However, these techniques are hampered by problems such as propagation of errors, which results in inferior estimates, and high communication overheads leading to poor scalability. In view of this, a novel Selective Iterative Multilateration (SIM) algorithm is described in this paper to improve the accuracy of location estimation in hop count-based localization schemes without incurring unnecessary overhead costs. New anchor nodes are selected judiciously such that their initial position estimates are sufficiently accurate. Also, such new anchor nodes are prevented from appearing in the same regions so that unnecessary overhead is kept to a minimum.