While learning-based Multi-View Stereo (MVS) excels in sparse-view reconstruction, refining its output with 3D Gaussian Splatting (3DGS) remains challenging. Excessive positional degrees of freedom (DoFs) in Gaussians often cause instability and geometric artifacts, sometimes distorting geometry to represent texture patterns. To address this issue, we propose GDoFS (Gaussian DoF Separation), a strategy that divides positional DoFs into two categories—image-plane-parallel and ray-aligned—based on their uncertainty. For each category, GDoFS introduces tailored optimization techniques, including bounded offsets for low-uncertainty DoFs and a visibility-guided loss for ray-aligned DoFs. Experiments on standard benchmarks demonstrate that GDoFS effectively mitigates geometric artifacts and produces reconstructions that are both visually coherent and structurally accurate.

Mehrab Tanjim, Yeonjun In, Xiang Chen, Victor Bursztyn, Ryan A. Rossi, Sungchul Kim, Guang-Jie Ren, Vaishnavi Muppala, Shun Jiang, Yongsung Kim, Chanyoung Park. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Reviewing the literature to understand relevant threads of past work is a critical part of research and vehicle for learning. However, as the scientific literature grows the challenges for users to find and make sense of the many different threads of research grow as well. Previous work has helped scholars to find and group papers with citation information or textual similarity using standalone tools or overview visualizations. Instead, in this work we explore a tool integrated into users’ reading process that helps them with leveraging authors’ existing summarization of threads, typically in introduction or related work sections, in order to situate their own work’s contributions. To explore this we developed a prototype that supports efficient extraction and organization of threads along with supporting evidence as scientists read research articles. The system then recommends further relevant articles based on user-created threads. We evaluate the system in a lab study and find that it helps scientists to follow and curate research threads without breaking out of their flow of reading, collect relevant papers and clips, and discover interesting new articles to further grow threads.

Consumers conducting comparison shopping, researchers making sense of competitive space, and developers looking for code snippets online all face the challenge of capturing the information they find for later use without interrupting their current flow. In addition, during many learning and exploration tasks, people need to externalize their mental context, such as estimating how urgent a topic is to follow up on, or rating a piece of evidence as a “pro” or “con,” which helps scaffold subsequent deeper exploration. However, current approaches incur a high cost, often requiring users to select, copy, context switch, paste, and annotate information in a separate document without offering specific affordances that capture their mental context. In this work, we explore a new interaction technique called “wiggling,” which can be used to fluidly collect, organize, and rate information during early sensemaking stages with a single gesture. Wiggling involves rapid back-and-forth movements of a pointer or up-and-down scrolling on a smartphone, which can indicate the information to be collected and its valence, using a single, light-weight gesture that does not interfere with other interactions that are already available. Through implementation and user evaluation, we found that wiggling helped participants accurately collect information and encode their mental context with a 58% reduction in operational cost while being 24% faster compared to a common baseline.

Tabs have become integral to browsing the Web yet have changed little since their introduction nearly 20 years ago. In contrast, the internet has gone through dramatic changes, with users increasingly moving from navigating to websites to exploring information across many sources to support online sensemaking. This paper investigates how tabs today are overloaded with a diverse set of functionalities and issues users face when managing them. We interviewed ten information workers asking about their tab management strategies and walk through each open tab on their work computers four times over two weeks. We uncovered competing pressures pushing for keeping tabs open (ranging from interaction to emotional costs) versus pushing for closing them (such as limited attention and resources). We then surveyed 103 participants to estimate the frequencies of these pressures at scale. Finally, we developed design implications for future browser interfaces that can better support managing these pressures.

Despite the increasing complexity and scale of people’s online activities, browser interfaces have stayed largely the same since tabs were introduced in major browsers nearly 20 years ago. The gap between simple tab-based browser interfaces and the complexity of users’ tasks can lead to serious adverse effects – commonly referred to as “tab overload.” This paper introduces a Chrome extension called Tabs.do, which explores bringing a task-centric approach to the browser, helping users to group their tabs into tasks and then organize, prioritize, and switch between those tasks fluidly. To lower the cost of importing, Tabs.do uses machine learning to make intelligent suggestions for grouping users’ open tabs into task bundles by exploiting behavioral and semantic features. We conducted a field deployment study where participants used Tabs.do with their real-life tasks in the wild, and showed that Tabs.do can decrease tab clutter, enabled users to create rich task structures with lightweight interactions, and allowed participants to context-switch among tasks more efficiently.

People receive dozens, or hundreds, of notifications per day and each notification poses some risk of accidental information disclosure in the presence of others; onlookers may see notifications on a mobile phone lock screen, on the periphery of a desktop or laptop display. We quantify the prevalence of these accidental disclosures in the context of email notifications, and we study people's relevant preferences and concerns. Our results are compiled from a retrospective survey of 131 respondents, and a contextual-labeling study where 169 participants labeled 1,040 meeting-email pairs. We find that, for 53% of people, at least 1 in 10 email notifications poses an information disclosure risk, and the real or perceived severity of these risks depend both on user characteristics and the meeting or email attributes. We conclude by exploring machine learning for predicting people's comfort levels, and we present implications for the design of future social-context aware notification systems.

Participatory sensing systems in which people actively participate in the data collection process must account for both the needs of data contributors and the data collection goals. Existing approaches tend to emphasize one or the other, with opportunistic and directed approaches making opposing tradeoffs between providing convenient opportunities for contributors and collecting high-fidelity data. This paper explores a new, hybrid approach, in which collected data-even if low-fidelity initially-can provide useful information to data contributors and inspire further contributions. We realize this approach with 4X, a multi-stage data collection framework that first collects data opportunistically by requesting contributions at specific locations along users' routes and then uses collected data to direct users to locations of interest to make additional contributions that build data fidelity and coverage. To study the efficacy of 4X, we implemented 4X into LES, an application for collecting information about campus locations and events. Results from two field deployments (N = 95, N = 18) show that the 4X framework created 34% more opportunities for contributing data without increasing disruption, and yielded 49% more data by directing users to locations of interest. Our results demonstrate the value and potential of multi-stage, dynamic data collection processes that draw on multiple sources of motivation for data, and how they can be used to better meet data collection goals as data becomes available while avoiding unnecessary disruption.

We consider the challenge of motivating and coordinating large numbers of people to contribute to solving local, communal problems through their existing routines. In order to design such "on-the-go crowdsourcing" systems, there is a need for mechanisms that can effectively coordinate contributions to address problem solving needs in the physical world while leveraging people's existing mobility with minimal disruption. We thus introduce Hit-or-Wait, a general decision-theoretic mechanism that intelligently controls decisions over when to notify a person of a task, in ways that reason both about system needs across tasks and about a helper's changing patterns of mobility. Through simulations and a field study in the context of community-based lost-and-found, we demonstrate that using Hit-or-Wait enables a system to make efficient use of people's contributions with minimal disruptions to their routines without the need for explicit coordination. Interviews with field study participants further suggest that highlighting an individual's contribution to the global goal may help people value their contributions more.

Trust is essential for beginning and maintaining relationships online where assessing uncertainties and risks is difficult. While product delays have been shown to reduce trust in e-commerce settings, we understand little about the effect of delays on trust in the increasingly popular context of crowdfunding. In a mixed method study, we examine what factors influence backers' trust in crowdfunding when rewards are delayed. Based on in-depth interviews with crowdfunding participants, we found that a rich set of factors influenced backers' trust including backers' role identity and domain knowledge, backer's research on a creator's background, creators' communication during delays, and duration of delays. To better understand the factors affecting delays, we conducted a regression analysis with 4,089 delayed projects and found that the funding goal, number of backers, percent raised, number of reward levels, and creator's previous crowdfunding experience are associated with the duration of delay. We discuss design implications for managing delays and maintaining trust in crowdfunding.

Recent years have seen the growth of physical crowdsourcing systems (e.g., Uber; TaskRabbit) that motivate large numbers of people to provide new and improved physical tasking and delivery services on-demand. In these systems, opportunistically relying on people to make convenient contributions may lead to incomplete solutions, while directing people to do inconvenient tasks requires high incentives. To increase people's willingness to participate and reduce the need to incentivize participation, we study on-the-go crowdsourcing as an alternative approach that suggests tasks along people’s existing routes that are conveniently on their way. We explore as a first step in this paper the design of task notification policies that decide when, where, and to whom to suggest tasks. Situating our work in the context of practical problems such as package delivery and lost-and-found searches, we conducted controlled experiments that show how small changes in task notification policy can influence individual participation and actions in significant ways that in turn affect system outcomes. We discuss the implications of our findings on the design of future on-the-go crowdsourcing technologies and applications.

Citizen science and communitysensing applications allow everyday citizens to collect data about the physical world to benefit science and society. Yet despite successes, current approaches are still limited by the number of domain-interested volunteers who are willing and able to contribute useful data. In this paper we introduce habitsourcing, an alternative approach that harnesses the habit-building practices of millions of people to collect environmental data. To support the design and development of habitsourcing apps, we present (1) interaction techniques and design principles for sensing through actuation, a method for acquiring sensing data from cued interactions; and (2) ExperienceKit, an iOS library that makes it easy for developers to build and test habitsourcing applications. In two experiments, we show that our two proof-of-concept apps, ZenWalk and Zombies Interactive, compare favorably to their non-data collecting counterparts, and that we can effectively extract environmental data using simple detection techniques.

This paper introduces the idea of presenting physical tasks into people’s existing routine so that people can contribute small tasks on-the-go with minimal effort. We present two on-the-go crowdsourcing systems: Libero for package delivery, and CrowdFound for finding lost items. To encourage contributions, we introduce notification techniques that present task opportunities when potential helpers are likely to accept. To direct people to regions where help is most needed, we introduce techniques for tracking a person’s location within a task region and directing their attention based on task history. Evaluation studies demonstrate the feasibility of on-the-go crowdsourcing and investigate questions over the likelihood of task completion, the perceived cost of disruption, and the effectiveness of tracking and coordination.

In location-based social Q&A, the questions related to a local community (e.g. local services and places) are typically answered by local residents (i.e. who have the local knowledge). In this work, we wanted to deepen our understanding of the localness of knowledge sharing through investigating the topical and typological patterns related to the geographic characteristics, geographic locality of user activities, and motivations of local knowledge sharing. To this end, we analyzed a 12-month period Q&A dataset from Naver KiN "Here" and a supplementary survey dataset from 285 mobile users. Our results revealed several unique characteristics of location-based social Q&A. When compared with conventional social Q&A sites, Naver KiN "Here" had very different topical/typological patterns. Naver KiN "Here" exhibited a strong spatial locality where the answerers mostly had 1-3 spatial clusters of contributions, the topical distributions varied widely across different districts, and a typical cluster spanned a few neighboring districts. In addition, we uncovered unique motivators, e.g. ownership of local knowledge and sense of local community. The findings reported in the paper have significant implications for the design of Q&A systems, especially location-based social Q&A systems.