Search

Proceedings of Graphics Interface 2000, Montréal, Québec, Canada, Canada, 15 - 17 May 2000, 53-60, This paper describes a technique for using magnetic motion capture data to determine the joint parameters of an articulated hierarchy. This technique makes it possible to determine limb lengths, joint locations, and sensor placement for a human subject without external measurements. Instead, the joint parameters are inferred with high accuracy from the motion data acquired during the capture Session. The parameters are computed by performing a linear least squares fit of a rotary joint model to the input data. A hierarchical structure for the articulated model can also be determined in situations where the topology of the model is not known. Once the system topology and joint parameters have been recovered, the resulting model can be used to perform forward and inverse kinematic procedures. We present the results of using the algorithm on human motion capture data, as well as validation results obtained with data from a simulation and a wooden linkage of known dimensions.

Postural control is a constant re-establishment process for the maintenance of balance and stability. Asymmetric weight distribution (AWD), characterized by uneven leg loading, leads to increased instability, injury, and progressive deterioration of posture and gait. Postural self-correction is automatically affected by the human body in response to visual, vestibular, and proprioceptive sensory information. However, simultaneous cognitive loads can increase the demand for extra resources and require balance monitoring and correction techniques. We address these issues with a novel physiological feedback system that utilizes load sensors for AWD detection, and electrical muscle stimulation (EMS) for automatic correction and restoration of balance by affecting a counter-weight shift. In a user study involving 36 participants, we compare our automatic approach against two alternative feedback systems (Audio and Vibro-tactile). We find that our automatic approach delivered faster correction and outperformed alternative feedback mechanisms and perceived to be interesting, comfortable and a potential commercial product.

Monte Carlo rendering algorithms can generate high-quality images; however they need to sample many rays per pixel and thus are computationally expensive. In this paper, we present a method to speed up Monte Carlo rendering by significantly reducing the number of pixels that we need to sample rays for. Specifically, we develop a neural future frame synthesis method that quickly predicts future frames from frames that have already been rendered. In each future frame, there are pixels that cannot be predicted correctly from previous frames in challenging scenarios, such as quick camera motion, object motion, and large occlusion. Therefore, our method estimates a mask together with each future frame that indicates the subset of pixels that need ray samples to correct the prediction results. To train and evaluate our neural future frame synthesis method, we develop a large ray-tracing animation dataset. Our experiments show that our method can significantly reduce the number of pixels that we need to render while maintaining high rendering quality.

The Stack Overflow Q&A community has been frequently criticized for being a harsh, unfriendly environment. Despite numerous calls by the community to improve in this regard, prior work has shown that negative community dynamics continue to deter women, newcomers, and other marginalized groups from getting engaged. Social feedback can play a significant role in shaping community behaviour through group norm reinforcement and can, therefore, be employed as a tool to create more welcoming environments. With this in mind, in this paper we present the design and evaluation of a visible social feedback mechanism for inclusion in a Q&A platform like Stack Overflow. Through an exploratory interview study with 20 Stack Overflow members (10 men, 10 women), we explore users' perceptions of the mechanism's potential benefits and drawbacks. Our findings suggest that compared to the men in our study, the women were more open to additional social feedback on Stack Overflow, finding it a potential solution to make Stack Overflow more welcoming. Our interview findings also suggest that such a tool could be used to encourage newcomers and to allow users to show appreciation for supportive phrasing, complementing Stack Overflow's existing focus on feedback for technically accurate content.

The 2022 CHCCS Achievement Award from the Canadian Human Computer Communications Society is presented to Professor Pourang Irani (UBC Okanagan) for his impact to the field of human-computer interaction. His research interests sit at the cross-section of MobileHCI and Information Visualization, and lately include the design and evaluation of in-situ user interfacese for improve sense-making. CHCCS invites a publication by the award winner to be included in the proceedings, and this year we continue the tradition of an interview format rather than a formal paper. This permits a casual discussion of the research areas, insights, and contributions of the award winner. What follows is an edited transcript of a conversation between Pourang Irani and Xing-Dong (Associate Professor at Simon Fraser University) that took place on April 20th, 2022.

We report on the findings from a mixed-methods user study that explores some of the less-studied challenges in designing personal visualizations. We implemented an interface presenting visualizations of the personal data gathered as part of a prior study and conducted a think-aloud study (N=15) of participants' exploration of their respective data on the interface. We analyzed participant verbal reports and interactions to (i) corroborate the types of insights they gained with the insight types identified in the literature, (ii) identify contextual information recalled by the participants to interpret their data, and (iii) identify interface design choices that potentially hinder insight discovery. Our findings complement prior work and we present design directions for visualizations of personal data, including guidelines for providing baselines for comparison, providing higher explorability for more data-savvy users, and minimizing the need for reading even in more exploratory interfaces.

We take the well-established use of physical scale models in architecture and identify new opportunities for using them to interactively visualize and examine multiple streams of geospatial data. Overlaying, comparing, or integrating visualizations of complementary data sets in the same physical space is often challenging given the constraints of various data types and the limited design space of possible visual encodings. Our vision of "simultaneous worlds" uses physical models as a substrate upon which visualizations of multiple data streams can be dynamically and concurrently integrated. To explore the potential of this concept, we created three design explorations that use an illuminated campus model to integrate visualizations about building energy use, climate, and movement paths on a university campus. We use a research through design approach, documenting how our interdisciplinary collaborations with domain experts, students, and architects informed our designs. Based on our observations, we characterize the benefits of models for 1) situating visualizations, 2) composing visualizations, and 3) manipulating and authoring visualizations. Our work highlights the potential of physical models to support embodied exploration of spatial and non-spatial visualizations through fluid interactions.

We present a novel method to preprocess a reduced model, training a neural network to approximate the reduced model signed distance field using active learning technique. The trained neural network is used to evaluate the self-collision state as well as the self-collision handling during real time simulation. Our offline learning approach consists of two passes of learning. The first pass learning generates positive and negative point cloud which is used in the second pass learning to learn the signed distance field of reduced subspace. Unlike common fully supervised learning approaches, we make use of semi-supervised active learning technique in generating more informative samples for training, improving the convergence speed. We also propose methods to use the learned SDF function in real time self-collision detection and assemble it in the constraint Jacobian matrix to solve the self-collision.

Spatial augmented reality (SAR) displays digital content in a real environment in ways that are situated, peripheral, and viewable by multiple people. These capabilities change how embedded information visualizations are used, designed, and experienced. But a comprehensive design framework that captures the specific characteristics and parameters relevant to SAR is missing. We contribute a new design framework for leveraging context and surfaces in the environment for SAR visualizations. An accompanying design process shows how designers can apply the framework to generate and describe SAR visualizations. The framework captures how the user's intent, interaction, and six environmental and visualization factors can influence SAR visualizations. The potential of this design framework is illustrated through eighteen exemplar application scenarios and accompanying envisionment videos.

We present FaceUI, a novel strategy to access mid-air face-centered spatial interfaces with off-the-shelf smartphones. FaceUI uses the smartphone's front-facing camera to track the phone's mid-air position relative to the user's face. This self-contained tracking mechanism opens up new opportunities to enable mid-air interactions on off-the-shelf smartphones. We demonstrate one possibility that leverages the empty mid-air space in front of the user to accommodate virtual windows which the user can browse by moving the phone in the space in front of their face. We inform our implementation of FaceUI by first studying essential design factors, such as the comfortable face-to-phone distance range and appropriate viewing angles for browsing mid-air windows and visually accessing their content. After that, we compare users' performance with FaceUI to their performance when using a touch-based interface in an analytic task that requires browsing multiple windows. We find that FaceUI offers better performance than the traditional touch-based interface. We conclude with recommendations for the design and use of face-centered mid-air interfaces on smartphones.

Structured 2D patterns formed by the anisotropic distribution of arbitrary shapes are ubiquitous in nature and man-made environments. They may include both bounded and unbounded (extended fiber-like) shapes. In this work, we address the problem of interactively generating such patterns from a single exemplar sketched by a user. We build our solution on a new data structure, the Support Structure Hierarchy, computed from a multi-resolution analysis of the input exemplar, that encodes the main anisotropy directions at different scales as well as deviations from them. We propose an efficient method based on this structure to synthesize a similar distribution of shapes in an extended 2D domain. The user can also choose to hybridize multiple input exemplars by combining structural shapes extracted at different scales. As shown in a user study, our multi-scale solution generates structured shape-patterns that perceptually compete with state-of-the-art methods, whether learning-based or not. Moreover, our interactive solution, which requires no pre-calculation, fits well with the needs of an interactive authoring tool, where the user can not only sketch and extend 2D vector textures but also combine them seamlessly.

Cartoon effects described in animation principles are key to adding fluidity and style to animated characters. This paper extends the existing framework of Velocity Skinning to use skeletal acceleration, in addition to velocity, for cartoon-style effects on rigged characters. This Acceleration Skinning is able to produce a variety of cartoon effects from highly efficient closed-form deformers while remaining compatible with standard production pipelines for rigged characters. The paper showcases the introduction of the framework along with providing applications through three new deformers. Specifically, a followthrough effect is obtained from the combination of skeletal acceleration and velocity. Also, centrifugal stretch and centrifugal lift effects are introduced using rotational acceleration to model radial stretching and lifting. The paper also explores the application of effect-specific time filtering when combining deformations together allowing for more stylization and artist control over the results.

Many visualizations display large datasets in which it can be difficult for users to find (and re-find) specific items. In systems that provide highlighting tools (e.g., filtering or brushing), emphasized points can become "accidental landmarks" - visual anchors that help users remember locations that are near the emphasized points. Accidental landmarks could be useful (by aiding revisitation), but if users become dependent on them, removing or changing the highlighting could cause problems. We provide designers with information about these issues through two crowdsourced studies in which people learned a set of item locations (in visualizations with or without emphasized points); we then removed or changed the highlighting to see if performance suffered. In the first study, which used a simple grid of points, results showed that changing or removing emphasized points significantly impeded users' ability to re-find targets, but the highlighting did not improve performance during training. In the second study, which used a more complex scatterplot, we found that highlighting significantly improved performance during training, but that removing or changing the emphasis points only reduced refinding performance for a few target types. Our work demonstrates that visualization designers need to consider how transient visual effects such as emphasis can affect spatial learning and revisitation, and provides new knowledge about how visual features can affect performance.

2D cut paths for laser cutters are specific to material parameters (type, thickness), design parameters (overall size, desired joint fit), and laser settings. Changing any of these parameters - because, e.g., one wants to cut a design from a new material, or on a different machine - requires by-hand adjustment of the cut path to compensate. This adjustment is complicated by the fact that 2D path formats lack information about how high-level parameters influenced the design, leaving makers to infer the original author's intent when adjusting paths. We present metaSVG, an exchange format that includes joint parameters as metadata when exporting cut plans from parametric 3D CAD. This allows our software tool, metaSVG Print, to automatically adapt metaSVG cut plans to new sizes, materials, material thickness, and laser cutters. We demonstrate the effectiveness of metaSVG by augmenting existing designs with metadata and fabricating them with different materials and laser cutters.

Assessing older adults' motor control ability is crucial for early diagnosis of Parkinson's Disease (PD). In this paper, we investigate how to use deep learning to detect motor impairment of older adults via analyzing touchscreen typing data, which could result in the early diagnosis of PD. Our investigation shows that deep learning is promising in analyzing touchscreen typing data. Among the four deep learning models (LSTM, LSTM-CNN, CNN-LSTM, and 1D CNN), LSTM-CNN yields the best performance. On a 102-subject dataset [53], LSTM-CNN achieved an AUC of 0.95 and an F1-score of 0.90 in leave-one-out PD classification, improving the performance of previously used SVM method (AUC = 0.88, F1-score = 0.73) [53]. LSTM-CNN also performed well on an in-the-clinic typing dataset [27] (AUC = 0.86, F1-score = 0.86), and significantly improved the F1-score of the previously proposed 1D CNN method (AUC = 0.89, F1 = 0.80). The promising performance of the LSTM-CNN model can also be generalized to other touchscreen interactions including flick, drag, handwriting, and pinch. It achieved better performance than the previous SVM method [53]. Our research showed that deep learning is effective in detecting early PD's motor symptoms via analyzing smartphone interaction data, and the proposed LSTM-CNN model is a promising neural network structure for performing such analysis. Overall, our research advances the understanding of how to assess the motor control ability of older adults via smartphone interactions.

Medical operation planning requires high precision due to the health risks involved. In virtual reality-based osteotomy operation planning, medical professionals would like to use their hands instead of the controllers for osteotomy plane manipulation. However, using hands as an input method in virtual reality is challenging due to noisy hand tracking. We explored the perceptual structure of precise plane manipulation by conducting a controlled experiment to compare: (1) separable Push-Poke which dynamically selects plane manipulation parameters, (2) separable custom axis with Control-Display gain widget which provides user to select object manipulation parameters, (3) integral pinch-based direct manipulation. The perceptual structure of hand-based plane manipulation techniques is composed of (1) integral and fast direct manipulation and (2) separable slow technique that dynamically selects manipulation parameters.

High fidelity interactions between game characters and gaseous effects like smoke, fire and explosions are often neglected in realtime applications due to the high computational cost of simulating fluids. In addition, the pose of game characters is only known at runtime as it depends on input from the user. Thus simulation-suitable representations of surface geometry must be generated on the fly. Common approaches like conversion into signed distance fields are not feasible for high-resolution geometry due to the computational cost and the amount of memory required on the GPU to store these fields. We present a purely vortex particle based fluid model for games which is capable of resolving the collision between fluids and complex objects such as moving game characters in real time. To handle collisions, we use a collocation method which only require a set of disassociated particles stuck to collision surfaces. Contrary to most other vorticity based methods, we use a simple inversion free approach to obtain the collision velocity field on surfaces while at the same time avoiding the expensive pressure projection step associated with pressure based fluid solvers.

Users often rely on their collaborators to find relevant application features by observing them "over the shoulder" (OTS), usually in a synchronous co-located setting. However, as remote work settings have become more common, users can no longer rely on such in-person interaction with collaborators. Therefore, we investigate designs that help the user become aware of relevant features based on collaborators' feature usage habits. We created five design concepts as video prototypes which varied in five design dimensions: number of active collaborators, number of shared documents, specificity of comparison, user involvement, and goal of the feature awareness. Interviews (N=18) probing the design concepts indicate that collaborator-based feature awareness would be valuable for discovering novel features and producing a consistent style across the shared document, but some users may feel micromanaged or self-conscious. We conclude by reflecting on and expanding our design space and discussing future design directions supporting remote OTS learning.

This study introduces a new sketch-based hairstyle authoring interface for virtual characters captured using multi-view stereo. Individual hairs are typically modeled by interpolating guide hairs, which are downsampled hair strands representing the overall hair shapes. Unlike conventional geometric modeling, we incorporate the physical properties of hairs, such as gravity, collision, and bending resistance, in guide hair modeling. The use of physics-related shape optimization in the interface allows natural-looking 3D hair shapes to be modeled from minimal user specifications. We provide an interactive sketch tool that allows the user to specify hair shapes from multiple viewing angles. Our interface generates feedback regarding how much a designed sketch conforms to the laws of physics, enabling the user to strike a balance between artistic intention and physical naturalness. We further introduce an automatic sketch suggestion system based on the hair orientation obtained using image filtering.

It is challenging to design an intelligent virtual agent (IVA) that can point from the virtual to the real world and have users accurately recognize where it is pointing due to differences in perceptual cues between the two spaces. We designed an IVA with factors including: a situated display, appearance, and pointing gesture strategy to establish whether it is possible to have an IVA point accurately into the real world. With a real person pointing as a baseline, we performed an empirical study using our designed IVA and demonstrated that participants perceived the IVA's pointing to a physical location with comparable accuracy to a real person baseline. Specifically, we found that when the IVA is 230 cm away from the targets on average, the IVA outperformed the real person in the vertical dimension (10.22 cm, 28.8% less error) and achieved the same level of accuracy (11.58 cm) horizontally. Our integrated design choices provide a foundation for design factors to consider when designing IVAs for pointing and pave the way for future studies and systems in providing accurate pointing perception.

Step-by-step tutorials have emerged as a key means for learning complex software, but they are typically designed for individuals learning independently. In contrast, cooperative learning, where learners can help each other as they work, is a fundamental pedagogical technique with many established benefits. To extend these benefits to learning 3D-design software, this work investigates the design of remote cooperative software tutorial systems. We first conduct an observational study of pairs of participants working on 3D-design tutorials, which reveals a range of potential benefits, challenges, and strategies for cooperation. Our findings inform the design of TwoTorials, a cooperative stepby-step tutorial system that helps pairs of remote users establish shared 3D context, maintain awareness of each other's activities, and coordinate their efforts. A user study reveals several benefits to this approach, including enhanced cooperation between learners, reduced effort and mental demand, increased awareness of peer activities, and higher subjective engagement with the tutorial.

Many mobile users rely on autocorrection mechanisms during text entry on their smartphone. Previous studies investigated the effects of autocorrection mechanisms on typing speed and accuracy but did not explore the level of frustration and perceived mental workload often associated with autocorrection. Through a mixed-methods user study, we investigate the effect of autocorrection failures on increasing the user's frustration, mental and physical demand, performance, and effort in this paper. We identified that perceived mental and physical demand, and frustration are directly affected by autocorrection.

Non-expert annotators must select an appropriate label for an image when the annotation task is difficult. Then, it might be easier for an annotator to choose multiple "likely" labels than to select a single label. Multiple labels might be more informative in the training of a classifier because multiple labels can have the correct one, even when a single label is incorrect. We present DualLabel, an annotation tool that allows annotators to assign secondary labels to an image to simplify the annotation process and improve the classification accuracy of a trained model. A user study compared the proposed dual-label and traditional singlelabel approaches for an image annotation task. The results show that our dual-label approach reduces task completion time and improves classifier accuracy trained with the given labels.

Digital art tools allow children to express their creativity and can help them develop important skills. There are numerous software tutorials available to help teach and inspire digital art enthusiasts, however, most are authored for and by adults. Given that children are increasingly contributing online digital content, in this paper, we investigate a tutorial authoring design concept where children can capture their drawings and information on their process, with the long-term objective of allowing children to share both their creativity and their workflows with other children. Through participatory design sessions, prototyping, and an evaluation, we explore children's attitudes towards the creation of digital art tutorials, focusing on their perceived incentives to author such tutorials and how they feel about the concept of sharing their tutorials with other children. We also elicit reactions towards specific design elements. Our findings suggest important considerations for tools designed to motivate and support children's creation of digital art tutorials.

Assigning a label to difficult data requires a long time, particularly when non-expert annotators attempt to select the best possible label. However, there have been no detailed studies exploring a label selection style during annotation. This is very important and may affect the efficiency and quality of annotation. In this study, we explored the effects of labeling style on data annotation and machine learning. We conducted an empirical study comparing "quick labeling" and "careful labeling" styles in image-labeling tasks with three levels of difficulty. Additionally, we performed a machine learning experiment using labeled images from the two labeling styles. The results indicated that quick and careful labeling styles have both advantages and disadvantages in terms of annotation efficiency, label quality, and machine learning performance. Specifically, careful labeling improves label accuracy when the task is moderately difficult, whereas it is time-consuming when the task is easy or extremely difficult.

With the number of Alzheimer's patients reaching 5 million in 2014 according to the U.S. Center for Disease Control and Prevention, increasing emphasis has been placed on identifying and understanding its precursor condition, Mild Cognitive Impairment (MCI). MCI is characterized by subtle but abnormal cognitive decline and is challenging to detect without formal testing. Neuropsychologists use paper-and-pencil tests such as the Trail-Making Test (TMT) for diagnosis, and ongoing research places importance on high-granularity sketch data from digital TMTs. We present SmartStrokes, a digital TMT app designed to simulate the paper-and-pencil testing experience on a tablet and stylus. Our contribution frames the principles of digital sketch recognition and Human-Computer Interaction (HCI) into the existing neuropsychological test, outlining the creation of a pair of classification models that identify MCI on an individual segmented line basis. Such a per-line classification method which could provide localized sketching behavior indicative of MCI. We also present an interface for the digital TMT and a refinement of line segmentation algorithms from previous research to better distinguish between the actions that a participant takes when completing the exam.

Ambiguity, the state in which alternative interpretations are plausible or even desirable, is an inexorable part of complex sensemaking. Its challenges are compounded when analysis involves risk, is constrained, and needs to be shared with others. We report on several studies with avalanche forecasters that illuminated these challenges and identified how visualization designs can better support ambiguity. Like many complex analysis domains, avalanche forecasting relies on highly heterogeneous and incomplete data where the relevance and meaning of such data is context-sensitive, dependant on the knowledge and experiences of the observer, and mediated by the complexities of communication and collaboration. In this paper, we characterize challenges of ambiguous interpretation emerging from data, analytic processes, and collaboration and communication and describe several management strategies for ambiguity. Our findings suggest several visual analytics design approaches that explicitly address ambiguity in complex sensemaking around risk.

International audience; Text editing is important and at the core of most complex tasks, likewriting an email or browsing the web. Efficient and sophisticatedtechniques exist on desktops and touch devices, but are still under-explored for Augmented Reality Head Mounted Display (AR-HMD).Performing text selection, a necessary step before text editing, in ARdisplay commonly uses techniques such as hand-tracking, voice com-mands, eye/head-gaze, which are cumbersome and lack precision.In this paper, we explore the use of a smartphone as an input deviceto support text selection in AR-HMD because of its availability,familiarity, and social acceptability. We propose four eyes-free textselection techniques, all using a smartphone — continuous touch,discrete touch, spatial movement, and raycasting. We compare themin a user study where users have to select text at various granularitylevels. Our results suggest that continuous touch, in which we usedthe smartphone as a trackpad, outperforms the other three techniquesin terms of task completion time, accuracy, and user preference.

International audience; We report on a controlled user study in which we investigated and compared three selection techniques in discovering and traversing 3D objects in densely packed environments. We apply this to cell division history marking as required by plant biologists who study the development of embryos, for whom existing selection techniques do not work due to the occlusion and tight packing of the cells to be selected. We specifically compared a list-based technique with an additional 3D view, a 3D selection technique that relies on an exploded view, and a combination of both techniques. Our results indicate that the combination was most preferred. List selection has advantages for traversing cells, while we did not find differences for surface cells. Our participants appreciated the combination because it supports discovering 3D objects with the 3D explosion technique while using the lists to traverse 3D cells.

We present PaperPal, a wearable smartphone assistant which blind people can use to fill out paper forms independently. Unique features of PaperPal include: a novel 3D-printed attachment that transforms a conventional smartphone into a wearable device with adjustable camera angle; capability to work on both flat stationary tables and portable clipboards; real-time video tracking of pen and paper which is coupled to an interface that generates real-time audio read outs of the form's text content and instructions to guide the user to the form fields; and support for filling out these fields without signature guides. The paper primarily focuses on an essential aspect of PaperPal, namely an accessible design of the wearable elements of PaperPal and the design, implementation and evaluation of a novel user interface for the filling of paper forms by blind people. PaperPal distinguishes itself from a recent work on smartphone-based assistant for blind people for filling paper forms that requires the smartphone and the paper to be placed on a stationary desk, needs the signature guide for form filling, and has no audio read outs of the form's text content. PaperPal, whose design was informed by a separate wizard-of-oz study with blind participants, was evaluated with 8 blind users. Results indicate that they can fill out form fields at the correct locations with an accuracy reaching 96.7%.

Habitually poor posture can lead to repetitive strain injuries that lower an individual's quality of life and productivity. Slouching over computer screens and smart phones are common examples that leads to soreness, and stiffness in the neck, shoulders, upper and lower back regions. To help cultivate good postural habits, researchers have proposed slouch detection systems which alert users when their posture requires attention. However, such notifications are disruptive and can be easily ignored. We address these issues with a new physiological feedback system that uses inertial measurement unit sensors to detect slouching, and electrical muscle stimulation to automatically correct posture. In a user study involving 36 participants, we compare our automatic approach against two alternative feedback systems and through two unique contexts-text entry and gaming. We find that our approach was perceived to be more accurate, interesting, and outperforms alternative techniques in the gaming but not text entry scenario.

Our approach improves the accuracy of weighted blended orderindependent transparency, while remaining efficient and easy to implement. We extend the original algorithm to a layer-based approach, where the content of each layer is blended independently before compositing them globally. Hereby, we achieve a partial ordering but avoid explicit sorting of all elements. To ensure smooth transitions across layers, we introduce a new weighting function. Additionally, we propose several optimizations and demonstrate the method's effectiveness on various challenging scenes in terms of geometricand depth complexity. We achieve an error reduction more than an order of magnitude on average compared to weighted blended order-independent transparency for our test scenes.

Third-person is a popular perspective for video games, but virtual reality (VR) seems to be primarily experienced from a first-person point of view (POV). While a first-person POV generally offers the highest presence; a third-person POV allows users to see their avatar; which allows for a better bond, and the higher vantage point generally increases spatial awareness and navigation. Third-person locomotion is generally implemented using a controller or keyboard, with users often sitting down; an approach that is considered to offer a low presence and embodiment. We present a novel thirdperson locomotion method that enables a high avatar embodiment by integrating skeletal tracking with head-tilt based input to enable omnidirectional navigation beyond the confines of available tracking space. By interpreting movement relative to an avatar, the user will always keep facing the camera which optimizes skeletal tracking and keeps required instrumentation minimal (1 depth camera). A user study compares the performance, usability, VR sickness incidence and avatar embodiment of our method to using a controller for a navigation task that involves interacting with objects. Though a controller offers a higher performance and usability, our locomotion method offered a significantly higher avatar embodiment.

We present a method to produce stylized drawings from stereoscopic 3D (S3D) images. Taking advantage of the information provided by the disparity map, we extract object contours and determine their visibility. The discovered contours are stylized and warped to produce an S3D line drawing. Since the produced line drawing can be ambiguous in shape, we add stylized shading to provide monocular depth cues. We investigate using both consistently rendered shading and inconsistently rendered shading in order to determine the importance of lines and shading to depth perception.

We introduce a novel visual analytics approach for analyzing, understanding, and correcting neural machine translation. Our system supports users in automatically translating documents using neural machine translation and identifying and correcting possible erroneous translations. User corrections can then be used to fine-tune the neural machine translation model and automatically improve the whole document. While translation results of neural machine translation can be impressive, there are still many challenges such as overand under-translation, domain-specific terminology, and handling long sentences, making it necessary for users to verify translation results; our system aims at supporting users in this task. Our visual analytics approach combines several visualization techniques in an interactive system. A parallel coordinates plot with multiple metrics related to translation quality can be used to find, filter, and select translations that might contain errors. An interactive beam search visualization and graph visualization for attention weights can be used for post-editing and understanding machine-generated translations. The machine translation model is updated from user corrections to improve the translation quality of the whole document. We designed our approach for an LSTM-based translation model and extended it to also include the Transformer architecture. We show for representative examples possible mistranslations and how to use our system to deal with them. A user study revealed that many participants favor such a system over manual text-based translation, especially for translating large documents.

Most text entry techniques for smartwatches require repeated taps to enter one word, occupy most of the screen, or use layouts that are difficult to learn. Users are usually reluctant to use these techniques since the skills acquired in learning cannot be transferred to other devices. SwipeRing is a novel keyboard that arranges the QWERTY layout around the bezel of a smartwatch divided into seven zones to enable gesture typing. These zones are optimized for usability and to maintain similarities between the gestures drawn on a smartwatch and a virtual QWERTY to facilitate skill transfer. Its ring-shaped layout keeps most of the screen available. We compared SwipeRing with C-QWERTY that uses a similar layout but does not divide the keys into zones or optimize for skill transfer and target selection. In a study, SwipeRing yielded a 33% faster entry speed (16.67 WPM) and a 56% lower error rate than C-QWERTY.

There is a growing concern that smartphone usage in front of family or friends can be bothersome and even deteriorate relationships. We report on a survey examining smartphone usage behavior and problems that arise from overuse when partners (married couples, common-law relationships) are co-located. Results show that people have various expectations from their partner, and often feel frustrated when their partner uses a smartphone in front of them. Study participants also reported a lack of smartphone activity awareness that could help decide when or how to communicate expectations to the partner. This motivated us to develop an app, CoAware, for sharing smartphone activity-related information between partners. In a lab study with couples, we found that CoAware has the potential to improve smartphone activity awareness among co-located partners. In light of the study results, we suggest design strategies for sharing smartphone activity information among co-located partners.

In this work, we present the design and comparative evaluation of techniques for increasing awareness of out-of-view targets in virtual reality. We first compare two variants of SOUS-a technique that guides the user to out-of-view targets using circle cues in their peripheral vision-with the existing FlyingARrow technique, in which arrows fly from the user's central (foveal) vision toward the target. fSOUS, a variant with low visual salience, performed well in a simple environment but not in visually complex environments, while bSOUS, a visually salient variant, yielded faster target selection than both fSous and FlyingARrow across all environments. We then compare hybrid techniques in which aspects of SOUS relating to unobtrusiveness and visual persistence were reflected in design modifications made to FlyingARrow. Increasing persistence by adding trails to arrows improved performance but there were concerns about obtrusiveness, while other modifications yielded slower and less accurate target acquisition. Nevertheless, since fSOUS and bSOUS are exclusively for head-mounted display with wide field-of-view, FlyingARrow with trail can still be beneficial for devices with limited field-of-view.

The availability of voice-user interfaces (VUIs) has grown dramatically in recent years. As more capable systems invite higher expectations, the conversational interactions that VUIs support introduce ambiguity in accountability: a user's or system's obligation or willingness to be responsible for the outcome of user-delegated tasks. When misconstrued, impact ranges from inconvenience to deadly harm. This project explores how users' accountability perceptions and expectations can be managed in voice interaction with smart home appliances. To explore links between degree of automation, system accountability and user satisfaction, we identified key design factors for VUI design through an exploratory study, articulated them in video prototypes of four new VUI mechanisms showing a user commanding an advanced appliance and encountering a problem, and deployed them in a second study. We found that participants were more satisfied with automated systems, but also saw them as more accountable. Results suggest that this perceived system accountability was reduced in mechanisms that justified their recommendations. Our findings motivate the development of transitional mechanisms which gradually provide less guidance as users learn to operate the system.

Monte Carlo rendering algorithms are widely used to produce photorealistic computer graphics images. However, these algorithms need to sample a substantial amount of rays per pixel to enable proper global illumination and thus require an immense amount of computation. In this paper, we present a hybrid rendering method to speed up Monte Carlo rendering algorithms. Our method first generates two versions of a rendering: one at a low resolution with a high sample rate (LRHS) and the other at a high resolution with a low sample rate (HRLS). We then develop a deep convolutional neural network to fuse these two renderings into a high-quality image as if it were rendered at a high resolution with a high sample rate. Specifically, we formulate this fusion task as a super resolution problem that generates a high resolution rendering from a low resolution input (LRHS), assisted with the HRLS rendering. The HRLS rendering provides critical high frequency details which are difficult to recover from the LRHS for any super resolution methods. Our experiments show that our hybrid rendering algorithm is significantly faster than the state-of-the-art Monte Carlo denoising methods while rendering high-quality images when tested on both our own BCR dataset and the Gharbi dataset. \url{https://github.com/hqqxyy/msspl}, Comment: Graphic Interface 2021

Front matter (President's Welcoming Letter, Organization, Michael A. J. Sweeney Award 2021, Alain Fournier Award 2020, Bill Buxton Dissertation Award 2020, Achievement Award 2021, Achievement Award 2021).

International audience; Tracking in a cinematic way a moving target inside a 3D dynamic environment remains a challenging problem. This requires to simultaneously ensure a low computational cost, a good degree of reactivity and a high cinematic quality despite sudden changes. In this paper, we draw on the idea of Motion-Predictive Control to propose an efficient real-time camera tracking technique which ensures these properties. Our approach relies on the predicted motion of a target to create and evaluate a very large number of camera motions using hardware ray casting. Our evaluation of camera motions includes a range of cinematic properties such as distance to target, visibility, collision, smoothness and jitter. Experiments are conducted to display the benefits of the approach with relation to prior work.

We present Paper Forager, a web-based system which allows users to rapidly explore large collections of research documents. Our sample corpus uses 5,055 papers published at the ACM CHI and UIST conferences. Paper Forager provides a visually based browsing experience, allowing users to identify papers of interest based on their graphical appearance, in addition to providing traditional faceted search techniques. A cloud-based architecture stores the papers as multi-resolution images, giving users immediate access to reading individual pages of a paper, thus reducing the transaction cost between finding, scanning, and reading papers of interest. Initial user feedback sessions elicited positive subjective feedback, while a 24-month external deployment generated in-the-wild usage data which we analyze. Users of the system indicated that they would be enthusiastic to continue having access to the Paper Forager system in the future.

Computer users commonly use multi-step text entry methods on handheld devices and as alternative input methods for accessibility. These methods are also commonly used to enter text in languages with large alphabets or complex writing systems. These methods require performing multiple actions simultaneously (chorded) or in a specific sequence (constructive) to produce input. However, evaluating these methods is difficult since traditional performance metrics were designed explicitly for non-ambiguous, uni-step methods (e.g., QWERTY). They fail to capture the actual performance of a multi-step method and do not provide enough detail to aid in design improvements. We present three new action-level performance metrics: UnitER, UA, and UnitCX. They account for the error rate, accuracy, and complexity of multi-step chorded and constructive methods. They describe the multiple inputs that comprise typing a single character-action-level metrics observe actions performed to type one character, while conventional metrics look into the whole character. In a longitudinal study, we used these metrics to identify probable causes of slower text entry and input errors with two existing multi-step methods. Consequently, we suggest design changes to improve learnability and input execution.

Patient data visualization can help healthcare providers gain an overview of their patient's condition and assist in decision-making about the next steps on management and communication. We explore the acceptance and opinion of five different visualizations that can be used to summarize patient data, including a Text Summary, text and frequency-based Word Cloud, a Bar Graph, a time-based Line Graph and a newly developed Text Graph that combines text and time-based distribution. Results from a user study with 15 professional healthcare providers, 16 firstor secondyear medical students, and 17 third or greater year medical students show that most visualizations are useful in extracting patient information and are received positively by the users. In addition, Text Summary and Text Graph are rated to be the most useful visualizations in extracting patient health information.

International audience; Selecting targets accurately and quickly with eye-gaze input remains an open research question. In this paper, we introduce BayesGaze, a Bayesian approach of determining the selected target given an eye-gaze trajectory. This approach views each sampling point in an eye-gaze trajectory as a signal for selecting a target. It then uses the Bayes' theorem to calculate the posterior probability of selecting a target given a sampling point, and accumulates the posterior probabilities weighted by sampling interval to determine the selected target. The selection results are fed back to update the prior distribution of targets, which is modeled by a categorical distribution. Our investigation shows that BayesGaze improves target selection accuracy and speed over a dwell-based selection method, and the Center of Gravity Mapping (CM) method. Our research shows that both accumulating posterior and incorporating the prior are effective in improving the performance of eye-gaze based target selection.

Contour plots are widely used in geospatial data visualization as they provide natural interpretation of information across spatial scales. To compare a geospatial attribute against others, contour plots for the base attribute (e.g., elevation) are often overlaid, blended, or examined side by side with other attributes (e.g., temperature or pressure). Such visual inspection is challenging since overlay and color blending both clutter the visualization, and a side-by-side arrangement requires users to mentally integrate the information from different plots. Therefore, these approaches become less efficient as the number of attributes grows. In this paper we examine the fundamental question of whether the base contour lines, which are already present in the map space, can be leveraged to visualize how other attributes relate to the base attribute. We present five different designs for stylizing contour lines, and investigate their interpretability using three crowdsourced studies. Our first two studies examined how contour width and number of contour intervals affect interpretability, using synthetic datasets where we controlled the underlying data distribution. We then compared the designs in a third study that used both synthetic and real-world meteorological data. Our studies show the effectiveness of stylizing contour lines to enrich the understanding of how different attributes relate to the reference contour plot, reveal trade-offs among design parameters, and provide designers with important insights into the factors that influence interpretability.