Search

Objectives: To accurately differentiate clumpy artifacts from tophi with foot and ankle DECT. Methods and materials: In session 1, 108 clumpy artifacts from 35 patients and 130 tophi images from 25 patients were analyzed. Reviewers classified green pixelation according to anatomic location, shape (linear, stippled, angular, oval), and height and width ratio. In session 2, green pixelation confined to the tendon was evaluated (shape, height and width ratio, occupied area in the tendon, accompanied peritendinous green pixelation). Results: In session 1, while tophi were noted at various locations, almost all clumpy artifacts were located at the tendon (99%, p < 0.0001). Most clumpy artifacts were linear, stippled, and wide, while most tophi were angular and oval (p < 0.05). In session 2, the shape of green pixelation from clumpy artifacts and tophi was significantly different (p < 0.0001) and most clumpy artifacts occupied less than 50% of the tendon (p = 0.02), and most tophi were accompanied by peritendinous green pixelation (p < 0.0001). Univariant logistic regression showed that tophi were significantly correlated with peritendinous deposits, angular and oval shape, and more than 50% of the tendon (p < 0.05). Conclusion: Clumpy artifacts can be differentiated from tophi in DECT. Clumpy artifacts typically are located in the tendon with a linear or stippled shape, wide, and less than 50% of a tendon’s cross-section. Tophi, on the other hand, typically are oval, larger than 50% of the tendon’s cross-section, and associated with adjacent peritendinous green pixelation. Advances in knowledge: Clumpy artifacts can be differentiated from tophi in image findings by their location and shape.

Pixel Art je oblika digitalne umetnosti, ki temelji na upodabljanju razpoznavnih elementov z omejeno barvno paleto ter uporabo mreže, sestavljene iz polj v velikosti enega piksla. Je ena od začetnih oblik 2D računalniške grafike in deluje po principu upoštevanja omejitev zmogljivosti starejših procesorjev in katodnih zaslonov (CRT angl. cathode ray tube). Danes Pixel Art pridobiva na priljubljenosti zaradi popularnosti retro estetike in dejavnika nostalgije, prav tako pa je grafično nezahteven način razvijanja videoiger. Laiki ga pogosto enačijo s pikseliziranimi slikami, kar pa se zgodi zaradi zniževanja ločljivosti zaradi spreminjanja velikosti. To prepričanje je zmotno, saj je Pixel Art samostojna oblika digitalne umetnosti, ki zahteva precejšnjo količino vloženega časa in truda ter pri kateri se je treba ravnati po določenih smernicah. Diplomsko delo se ukvarja s tehniko Pixel Art ilustracije, pikselizacijo bitne oziroma rastrske grafike in razlikovanjem med pojmoma. V teoretičnem delu so razloženi principi računalniške grafike, digitalne umetnosti in sloga Pixel Art. Predstavljen je tudi pregled zgodovinskega razvoja računalniške grafike, saj je ta neposredno povezan s Pixel Art ilustracijo sodobnega časa. V eksperimentalnem delu so bile ustvarjene ilustracije v Pixel Art slogu in izdelane imitacije Pixel Art sloga, ki so bile analizirane in primerjane. V raziskavi diplomskega dela je bilo ugotovljeno, da se ilustracije od imitacij Pixel Art sloga vsekakor razlikujejo, a je opaznost razlike odvisna od upodobljene podobe. Izkazalo se je tudi, da najboljše približke Pixel Art slogu daje interpolacija najbližjega soseda. Pixel Art is a form of digital art based on the representation of recognisable elements with a limited colour palette and the use of a grid consisting of pixel-sized fields. It is one of the earlier forms of 2D computer graphics and is based on the principle of working with the limitations of older computer units and CRT monitors. Today, Pixel Art is growing in popularity due to its retro aesthetic and nostalgia factor, as well as the fact that it is a graphically undemanding way of developing videogames. Pixel Art is often equated with the pixelation of images that results from low resolution due to downsizing. However, this is wrong – Pixel Art is an independent form of digital art that requires a considerable amount of time and work and for which certain criteria must be met. This thesis deals with Pixel Art, bit-graphic pixelation and the distinction between the terms. The theoretical part explains the basics of computer graphics, digital art and Pixel Art style. It also covers the historical development of computer graphics, as it is directly related to today’s Pixel Art illustration. In the experimental part, Pixel Art style illustrations and imitated Pixel Art graphics were created. The results were compared and evaluated. The thesis investigation revealed that there is indeed a noticeable difference between Pixel Art illustrations and Pixel Art imitations, however, the visibility of the dissimilarity varies depending on the image depicted. It was also found that the best approximation to the Pixel Art style was achieved by Nearest Neighbour interpolation.

We propose a method of searching for radio sources exhibiting the Sunyaev-Zeldovich effect in the multi-frequency emission maps from the Planck mission data using a convolutional neural network. A catalog for recognizing radio sources is compiled using the GLESP pixelation scheme at the frequencies of 100, 143, 217, 353, and 545 GHz. The quality of the proposed approach is evaluated and the quality of the dependence of model data on the S/N ratio is estimated. We show that the presented neural network approach allows the detection of sources with the Sunyaev-Zeldovich effect. The proposed method can be used to find the most likely galaxy cluster candidates at large redshifts., 18 pages, 7 figures, 1 tabel

InGaN-based blue light-emitting diodes (LEDs), with their high efficiency and brightness, are entering the display industry. However, a significant gap remains between the expectation of highly efficient light sources and their experimental realization into tiny pixels for ultrahigh-density displays for augmented reality. Herein, we report using tailored ion implantation (TIIP) to fabricate highly efficient, electrically-driven pixelated InGaN micro-LEDs (μLEDs) at the mid-submicrometre scale (line/space of 0.5/0.5 μm), corresponding to 8,500 pixels per inch (ppi) (RGB). Creating a laterally confined non-radiative region around each pixel with a controlled amount of mobile vacancies, TIIP pixelation produces relatively invariant luminance, and high pixel distinctiveness, at submicrometre-sized pixels. Moreover, with the incomparable integration capability of TIIP pixelation due to its planar geometry, we demonstrate 2,000 ppi μLED displays with monolithically integrated thin-film transistor pixel circuits, and 5,000 ppi compatible core technologies. We expect that the demonstrated method will pave the way toward high-performance μLED displays for seamless augmented-reality glasses. Submicrometre-sized InGaN-based light-emitting diodes are fabricated by tailored ion implantation. The devices are free from electrical leakage and show a luminance of 7,440 nit at 4.9 A cm−2 even at the line/space scale of 0.5/0.5 μm (= 8,500 ppi).

The review considers the contribution of the results of measuring fluctuations of relic cosmic microwave background (CMB) radiation to cosmological model and the latest results of observational cosmology. We discussed the basics of the modern cosmological model, the stages of obtaining information from observational data of the Planck space observatory, namely, analysis of time series of observations, construction procedures maps, pixelation, separation of background components, calculation of the angular power spectrum, and determination of cosmological parameters.

The study of visual memory is typically concerned with an image's content: How well, and with what precision, we can recall which objects, people, or features we have seen in the past. But images also vary in their quality: The same object or scene may appear in an image that is sharp and highly resolved, or it may appear in an image that is blurry and faded. How do we remember those properties? Here six experiments demonstrate a new phenomenon of "vividness extension": a tendency to (mis)remember images as though they are "enhanced" versions of themselves - that is, sharper and higher quality than they actually appeared at the time of encoding. Subjects briefly saw images of scenes that varied in how blurry they were, and then adjusted a new image to be as blurry as the original. Unlike an old photograph that fades and blurs, subjects misremembered scenes as more vivid (i.e., less blurry) than those scenes had actually appeared moments earlier. Follow-up experiments extended this phenomenon to saturation and pixelation - with subjects recalling scenes as more colorful and resolved - and ruled out various forms of response bias. We suggest that memory misrepresents the quality of what we have seen, such that the world is remembered as more vivid than it is.

Purpose: This study aims to assess the novel and innovative method of Safe Eye Examination (SEE) technique using the model eye for the purpose of teaching, training, and resident examination. Methods: A questionnaire-based study (Descriptive Data) with 53 participants, including ophthalmology residents, fellows in various subspecialties, and trainee optometrists was used. In our study, we used the Reti Eye model, but instead of the usual retina template sheet, we used prominent pathological fundus photographs loaded into the model eye. The study participants were asked to view prominent pathological fundus images printed on a matte finish paper. A circular image was cut and was placed in the Reti Eye model. The candidates were made to perform indirect ophthalmoscopy with a + 20D lens and to fill up a Google image assessment scale questionnaire with characteristics, including pixelation, sharpness, contrast, reflexes, blotchy appearance, and diagnostic confidence, which were then analyzed and depicted. Association between categorical variables was analyzed using Fisher exact test and Chi-square test. A P value of less than 0.05 was considered statistically significant. All data were analyzed with a statistical software package (SPSS, Version 16.0 for Windows). Results: The number of positive responses (>90%) obtained for the pixelation, sharpness, contrast, reflexes, blotchy appearance, and diagnostic confidence of the image viewed were statistically more significant than the negative responses (P < 0.05). Conclusion: The SEE technique of using the model eye can be incorporated for teaching, training, and skill assessment in the examinations in these difficult times of COVID-19 (coronavirus disease 2019) pandemic.

To date, the privacy-protection intended pixelation tasks are still labor-intensive and yet to be studied. With the prevailing of video live streaming, establishing an online face pixelation mechanism during streaming is an urgency. In this paper, we develop a new method called Face Pixelation in Video Live Streaming (FPVLS) to generate automatic personal privacy filtering during unconstrained streaming activities. Simply applying multi-face trackers will encounter problems in target drifting, computing efficiency, and over-pixelation. Therefore, for fast and accurate pixelation of irrelevant people's faces, FPVLS is organized in a frame-to-video structure of two core stages. On individual frames, FPVLS utilizes image-based face detection and embedding networks to yield face vectors. In the raw trajectories generation stage, the proposed Positioned Incremental Affinity Propagation (PIAP) clustering algorithm leverages face vectors and positioned information to quickly associate the same person's faces across frames. Such frame-wise accumulated raw trajectories are likely to be intermittent and unreliable on video level. Hence, we further introduce the trajectory refinement stage that merges a proposal network with the two-sample test based on the Empirical Likelihood Ratio (ELR) statistic to refine the raw trajectories. A Gaussian filter is laid on the refined trajectories for final pixelation. On the video live streaming dataset we collected, FPVLS obtains satisfying accuracy, real-time efficiency, and contains the over-pixelation problems.

Hyperspectral PRISMA images are new and have not yet been evaluated for their ability to detect marine plastic litter. The hyperspectral PRISMA images have a fine spectral resolution, however, their spatial resolution is not high enough to enable the discrimination of small plastic objects in the ocean. Pansharpening with the panchromatic data enhances their spatial resolution and makes their detection capabilities a technological challenge. This study exploits, for the first time, the potential of using satellite hyperspectral data in detecting small-sized marine plastic litter. Controlled experiments with plastic targets of various sizes constructed from several materials have been conducted. The required pre-processing steps have been defined and 13 pansharpening methods have been applied and evaluated for their ability to spectrally discriminate plastics from water. Among them, the PCA-based substitution efficiently separates plastic spectra from water without producing duplicate edges, or pixelation. Plastic targets with size equivalent to 8% of the original hyperspectral image pixel coverage are easily detected. The same targets can also be observed in the panchromatic image, however, they cannot be detected solely by the panchromatic information as they are confused with other appearances. Exploiting spectra derived from the pan-sharpened hyperspectral images, an index combining methodology has been developed, which enables the detection of plastic objects. Although spectra of plastic materials present similarities with water spectra, some spectral characteristics can be utilized for producing marine plastic litter indexes. Based on these indexes, the index combining methodology has successfully detected the plastic targets and differentiated them from other materials.

In recent years, organic electronic materials have been shown to be a promising tool, even transplanted in vivo, for transducing light stimuli to non-functioning retinas. Here we developed a bio-hybrid optoelectronic device consisting of patterned organic polymer semiconductors interfaced with an electrolyte solution in a closed sandwich architecture in order to study the photo-response of photosensitive semiconducting layers or patterns in an environment imitating biological extracellular fluids. We demonstrate an artificial retina model composed of on an array of 42,100 pixels made of three different conjugated polymers via inkjet printing with 110 pixels/mm2 packing density. Photo-sensing through three-colour pixelation allows to resolve incoming light spectrally and spatially. The compact colour sensitive optoelectronic device represents an easy-to-handle photosensitive platform for the study of the photo response of artificial retina systems.

The advent of single molecule microscopy has rev- olutionized biological investigations by providing a powerful tool for the study of intercellular and intracellular trafficking processes of protein molecules which was not available before through conventional microscopy. In practice, pixelated detectors are used to acquire the images of fluorescently labeled objects moving in cellular environments. Then, the acquired fluorescence microscopy images contain the numbers of the photons detected in each pixel, during an exposure time interval. Moreover, instead of having the exact locations of detection of the photons, we only know the pixel areas in which the photons impact the detector. These challenges make the analysis of single molecule trajectories, from pixelated images, a complex problem. Here, we investigate the effect of pixelation on the parameter estimation of single molecule trajectories. In particular, we develop a stochastic framework to calculate the maximum likelihood estimates of the parameters of a stochastic differential equation that describes the motion of the molecule in living cells. We also calculate the Fisher information matrix for this parameter estimation problem. The analytical results are complicated through the fact that the observation process in a microscope prohibits the use of standard Kalman filter type approaches. The analytical framework presented here is illustrated with examples of low photon count scenarios for which we rely on Monte Carlo methods to compute the associated probability distributions

A novel fast proximal scanning method, to the best of our knowledge, termed fiber-core-targeted scanning (FCTS), is proposed for illuminating individual fiber cores sequentially to remove the pixelation effect in fiber bundle (FB) imaging. FCTS is based on a galvanometer scanning system. Through a dynamic control of the scan trajectory and speed using the prior knowledge of fiber core positions, FCTS experimentally verifies a precise sequential delivery of laser pulses into fiber cores at a maximal speed of 45,000 cores per second. By applying FCTS on a FB-based photoacoustic forward-imaging probe, the results demonstrate that FCTS eliminates the pixelation effect and improves the imaging quality.

Optical proximity correction (OPC) is a widely used resolution enhancement technique (RET) in optical lithography to improve the image fidelity and process robustness. The efficiency of OPC is very important, especially for full-chip modification with complicated circuit layout in advanced technology nodes. An efficient OPC method based on virtual edge and mask pixelation with two-phase sampling is proposed in this paper. All kinds of imaging distortions are classified into two categories of imaging anomalies, the inward shrinkage anomaly and the outward extension anomaly. The imaging anomalies are detected around the corners and along the boundaries of the mask features with several anomaly detection templates. Virtual edges are adaptively generated according to the local imaging anomalies. The virtual edges are shifted to adjust the distribution of transparent regions on the mask and modify the local imaging anomalies. Several constraints and strategies are applied for efficient modifications and global control of the contour fidelity. In addition, the diffraction-limited property of the imaging system is fully utilized to separate the imaging evaluations at a coarse sampling level and the mask modifications at a fine sampling level, through the mask pixelation with two-phase sampling. It accelerates the imaging evaluations and guarantees the modification resolution as well. Simulations and comparisons demonstrate the superior modification efficiency of the proposed method.

HTTP-based Adaptive Streaming (HAS) is the dominant Internet video streaming application. One specific HAS approach, Dynamic Adaptive Streaming over HTTP (DASH), is of particular interest, as it is a widely deployed, standardized implementation. Prior academic research has focused on networking and protocol issues, and has contributed an accepted understanding of the performance and possible performance issues in large deployment scenarios. Our work extends the current understanding of HAS by focusing directly on the impacts of choice of the video quality adaptation algorithm on end-user perceived quality. In congested network scenarios, the details of the adaptation algorithm determine the amount of bandwidth consumed by the application as well as the quality of the rendered video stream. HAS will lead to user-perceived changes in video quality due to intentional changes in quality video segments, or unintentional perceived quality impairments caused by video decoder artifacts such as pixelation, stutters, or short or long stalls in the rendered video when the playback buffer becomes empty. The HAS adaptation algorithm attempts to find the optimal solution to mitigate the conflict between avoiding buffer stalls and maximizing video quality. In this article, we present results from a user study that was designed to provide insights into “best practice guidelines” for a HAS adaptation algorithm. Our findings suggest that a buffer-based strategy might provide a better experience under higher network impairment conditions. For the two network scenarios considered, the buffer-based strategy is effective in avoiding stalls but does so at the cost of reduced video quality. However, the buffer-based strategy does yield a lower number of quality switches as a result of infrequent bitrate adaptations. Participants in buffer-based strategy do notice the drop in video quality causing a decrease in perceived QoE, but the perceived levels of video quality, viewer frustration, and opinions of video clarity and distortion are significantly worse due to artifacts such as stalls in capacity-based strategy. The capacity-based strategy tries to provide the highest video quality possible but produces many more artifacts during playback. The results suggest that player video quality has more of an impact on perceived quality when stalls are infrequent. The study methodology also contributes a unique method for gathering continuous quantitative subjective measure of user perceived quality using a Wii remote.

Nowadays, in situ analysis attracts the interests and becomes one of the main purposes in analytical chemistry. Design of portable analysis devices facilitates reaching this goal. An ideal analysis system contains different parts enabling extraction, detection and quantification of target analytes. Preparation of a portable quantification approach is a bottle neck in such system creation. Common lab analysis instruments do not have the transportation ability and using these facilities limits the complete in situ analysis. Smartphones are the modern life phenomena and their usage becomes more widespread, every day. Their abilities and features are also swiftly developed. There are several strategies making the smartphone a suitable quantifier. This paper provides an overview of the currently applications of smartphones in analytical chemistry. Different applications of smartphones including optical detection (colorimetric, fluorescence, chemiluminescence, bioluminescence, and photoluminescence detections, pixelation as well as label-free detection), electrochemical detection, barcode reading, chemometric applications and smartphone imaging with fluorescence microscopy were classified and advantages and disadvantages of each approach were investigated. This modern common item could be a new part of analytical chemistry.

This paper aims to introduce the convolutional neural network (CNN) into modeling driver’s visual field (VF) using mobile laser scanning (MLS) data. A new solution that incorporates CNN is proposed to tackle the issues of inefficiency and inadequate manners of visualization in existing methods. The method operates along vehicle trajectory recorded in MLS data. For any driver position, the initial VF is defined as a fan-shaped area originating at the driver’s viewpoint. Within the initial VF, numerous virtual line-of-sights (LOS) are emitted from the viewpoint. Given an object point in any LOS, three-dimensional (3D) MLS points that may affect its visibility are converted to two-dimensional (2D) points using the cylindrical perspective projection. 2D points on the projective surface are then transformed into a binary image via the Pixelation procedure. Fed with the generated image, the CNN which is trained based on 789,500 data will classify the visibility as: 0-visible or 1-invisible. The location of the obstacle that blocks the driver’s view along each LOS is detected with a combination of the trained CNN and the bisection method. With all positions of obstacles determined, the final VF is established. Through comparisons with a state-of-the-art method, the CNN-based method shows remarkable efficiency, which facilitates either VF modeling at a single position or successive VF analyses along the vehicle path. A case study is also presented to show the improved manners of data visualization implemented in the developed method: (1) 3D viewshed, (2) sight distance curve, and (3) the driver’s perspective image with obstacles spotlighted.

Prediction is essential for the efficiency of many cognitive processes; however, this process is not always perfect. Predictive coding theory suggests that the brain generates and updates a prediction to respond to an upcoming event. Although an electrophysiological index of prediction, the stimulus preceding negativity (SPN), has been reported, it remains unknown whether the SPN reflects the prediction accuracy, or whether it is associated with the prediction error, which corresponds to a mismatch between a prediction and an actual input. Thus, the present study aimed to investigate this question using electroencephalography (EEG). Participants were asked to predict the original pictures from pictures that had undergone different levels of pixelation. The SPN amplitude was affected by the level of pixelation and correlated with the subjective evaluation of the prediction accuracy. Furthermore, late positive components (LPC) were negatively correlated with SPN. These results suggest that the amplitude of SPN reflects the prediction accuracy; more accurate prediction increases the SPN and reduces the prediction error, resulting in reduced LPC amplitudes.

In recent years, the development of computational imaging technology provides a new method for the realization of non-scanning super-resolution imaging. In this paper, a pixel super-resolution algorithm based on Fourier ptychographic technology is proposed, and the corresponding integrated and systematic programmable aperture coded super-resolution imaging system is constructed. By modulating the intensity with the coded aperture mask, utilizing different system point spread functions to obtain multiple samples of the original scene, and finally adopting sparse optimization iterative algorithm to reconstruct the original image, the result of super- resolution imaging is more than 3.5 times of Nyquist sampling frequency. In this tutorial, the proposed new super-resolution photoelectric imaging technology innovatively adopts the approach of coded aperture to realize image super-resolution imaging and effectively solve image pixelation. High-resolution images beyond the spatial resolution of the detector are obtained without any physical moving device or scanning mechanism. Compared with the traditional micro-scanning technology, it not only improves the reliability and stability of the system but also greatly reduces the cost and volume weight of the system.

With the prevailing of live video streaming, establishing an online pixelation method for privacy-sensitive objects is an urgency. Caused by the inaccurate detection of privacy-sensitive objects, simply migrating the tracking-by-detection structure into the online form will incur problems in target initialization, drifting, and over-pixelation. To cope with the inevitable but impacting detection issue, we propose a novel Privacy-sensitive Objects Pixelation (PsOP) framework for automatic personal privacy filtering during live video streaming. Leveraging pre-trained detection networks, our PsOP is extendable to any potential privacy-sensitive objects pixelation. Employing the embedding networks and the proposed Positioned Incremental Affinity Propagation (PIAP) clustering algorithm as the backbone, our PsOP unifies the pixelation of discriminating and indiscriminating pixelation objects through trajectories generation. In addition to the pixelation accuracy boosting, experiments on the streaming video data we built show that the proposed PsOP can significantly reduce the over-pixelation ratio in privacy-sensitive object pixelation.

Binarization is a key step in document analysis and archiving. The state-of-the-art models for document image binarization are variants of the encoder-and-decoder architecture, such as fully convolutional network (FCN) and U-Net. Despite their success, they still suffer from two challenges: (1) max-pooling or strided convolution reduces the spatial resolution of the intermediate feature maps, which may lead to information loss, and (2) interpolation or transposed convolution attempts to restore the feature maps to the desired spatial resolution, which may also result in pixelation. To overcome these two limitations, we propose a fully dilated convolutional network, termed FD-Net, using atrous convolutions instead of downsampling or upsampling operations. We have conducted extensive experiments on the recent DIBCO (document image binarization competition) and H-DIBCO (handwritten document image binarization competition) benchmark datasets. The experimental results show that our proposed FD-Net outperforms other state-of-the-art techniques by a large margin. The source code and pre-trained models are publicly available at https://github.com/beargolden/FD-Net.

Incom Inc. Large Area Picosecond Photodetector (LAPPD) 38 has been tested at Jefferson Lab to identify single-photoelectron signals to assess the potential of this type of device for future applications in Cherenkov light detection. Single-photoelectron signals were clearly detected if a tight collimation of photons impinging on the photocathode was used compared to the pixelation of the charge collection signal board., Comment: 17 pages, 15 figures

Main targets of this activity research are the making and the optimization of new detectors by means of the Systems Engineering methods. With the observation of the gravitational wave event of August 17th, 2017 and then with those of the extragalactic neutrino of September 22nd, the Multimessenger Astrophysics era began. It is a new way of exploring the Universe, powered by globally coordinated observations of several experiments. So, new X and gamma rays’ detectors solutions are needed in order to provide competitive results in the energy range 10 keV–10 meV. Here is briefly described how the Systems Engineering can improve the development of the proposal of a new technique: The Crystal Eye, a wide field of view detector with a good spatial resolution obtained thanks to a high pixelation.

In situ analysis is an important strategy that has attracted important attention from the scientific community in different fields of science. However, one of the main bottlenecks of this approach is associated with the absence of analysis tools that allow such analysis with significant accuracy and precision. In this sense, several smartphone applications, including optical detection (colorimetric, fluorescence, chemiluminescence, bioluminescence and photoluminescence, pixelation and unlabeled detection), electrochemical detection, smartphone image for minimally invasive neurosurgery with microscopy have been observed in the literature as an alternative technique of analysis, highlighting the potential of this tool for in situ performance analysis. However, many challenges are still observed using smartphone technology as analysis tools. In this chapter, we describe the main smartphone applications in the areas of chemistry, clinical, biological, environmental and forensics, highlighting the instrumentation used, advantages and limitation of each approach investigated.

In this work, the effect of randomly distributed stuck-at faults (SAFs) in memristive cross-point array (CPA)-based single and multi-layer perceptrons (SLPs and MLPs, respectively) intended for pattern recognition tasks is investigated by means of realistic SPICE simulations. The quasi-static memdiode model (QMM) is considered here for the modelling of the synaptic weights implemented with memristors. Following the standard memristive approach, the QMM comprises two coupled equations, one for the electron transport based on the double-diode equation with a single series resistance and a second equation for the internal memory state of the device based on the so-called logistic hysteron. By modifying the state parameter in the current-voltage characteristic, SAFs of different severeness are simulated and the final outcome is analysed. Supervised ex-situ training and two well-known image datasets involving hand-written digits and human faces are employed to assess the inference accuracy of the SLP as a function of the faulty device ratio. The roles played by the memristor’s electrical parameters, line resistance, mapping strategy, image pixelation, and fault type (stuck-at-ON or stuck-at-OFF) on the CPA performance are statistically analysed following a Monte-Carlo approach. Three different re-mapping schemes to help mitigate the effect of the SAFs in the SLP inference phase are thoroughly investigated.

Human pose estimation is a fundamental yet challenging task in computer vision, which aims at localizing human anatomical keypoints. However, unlike human vision that is robust to various data corruptions such as blur and pixelation, current pose estimators are easily confused by these corruptions. This work comprehensively studies and addresses this problem by building rigorous robust benchmarks, termed COCO-C, MPII-C, and OCHuman-C, to evaluate the weaknesses of current advanced pose estimators, and a new algorithm termed AdvMix is proposed to improve their robustness in different corruptions. Our work has several unique benefits. (1) AdvMix is model-agnostic and capable in a wide-spectrum of pose estimation models. (2) AdvMix consists of adversarial augmentation and knowledge distillation. Adversarial augmentation contains two neural network modules that are trained jointly and competitively in an adversarial manner, where a generator network mixes different corrupted images to confuse a pose estimator, improving the robustness of the pose estimator by learning from harder samples. To compensate for the noise patterns by adversarial augmentation, knowledge distillation is applied to transfer clean pose structure knowledge to the target pose estimator. (3) Extensive experiments show that AdvMix significantly increases the robustness of pose estimations across a wide range of corruptions, while maintaining accuracy on clean data in various challenging benchmark datasets., Comment: Accepted by CVPR 2021

The standard pipeline for many vision tasks uses a conventional camera to capture an image that is then passed to a digital processor for information extraction. In some deployments, such as private locations, the captured digital imagery contains sensitive information exposed to digital vulnerabilities such as spyware, Trojans, etc. However, in many applications, the full imagery is unnecessary for the vision task at hand. In this paper we propose an optical and analog system that preprocesses the light from the scene before it reaches the digital imager to destroy sensitive information. We explore analog and optical encodings consisting of easily implementable operations such as convolution, pooling, and quantization. We perform a case study to evaluate how such encodings can destroy face identity information while preserving enough information for face detection. The encoding parameters are learned via an alternating optimization scheme based on adversarial learning with deep neural networks. We name our system CAnOPIC (Camera with Analog and Optical Privacy-Integrating Computations) and show that it has better performance in terms of both privacy and utility than conventional optical privacy-enhancing methods such as blurring and pixelation.