Teleportation in Virtual Reality (VR) allows users to instantly move between locations, typically by aiming at a destination with a parabolic ray emitted from a controller and confirming the action with a button press. Recent advances in hand-tracking within Head-Mounted Displays (HMDs) have enabled researchers to explore hand-based teleportation, but these techniques often suffer from reduced pointing accuracy and increased effort. To address these challenges, we introduce NinjaPort, a novel, hand-based teleportation technique that allows users to control four virtual copies of their dominant hand with rays, each positioned at a different distance. Users can quickly select a desired teleportation distance by pinching a specific finger on their non-dominant hand, where each finger (e.g., index to pinky) maps to one of the four virtual hands. In our first user study, we explored design parameters to determine the optimal spatial arrangement of the virtual hands. Based on the findings, we refined NinjaPort and conducted a second study comparing it against a commercially available hand-based teleportation method and a research-based alternative. Our findings show that NinjaPort enables more accurate and faster teleportation, with less hand rotation. We conclude with design guidelines for creating effective and user-friendly hand-based VR teleportation techniques.

Medical image segmentation (MIS) is challenged by anatomical variability, ambiguous boundaries, and subtle textures, demanding an efficient balance between fine local details and global context. Existing architectures often suffer from suboptimal fusion of spatial and frequency-domain features, limiting their ability to capture richer structural and textural representations. To specifically address these challenges, we introduce FLoMo-Net, a modular MIS architecture organized around a principled route → select → refine → correct pipeline: (1) The Local–Global Mixture of Experts encoder adaptively routes features across specialized convolutional branches to capture scale-appropriate context; (2) The Dual-Attention Selective Aggregator then jointly selects informative channels and spatial regions using frequency-guided modulation; (3) At the bridge, the Frequency-Aware Multi-Scale Refinement module refines edges and textures through explicit low/high-frequency decomposition; (4) Finally, the False Positive/Negative Corrective Attention Module leverages uncertainty, derived from entropy and cosine dissimilarity, to produce a residual corrective mask that suppresses semantic drift and improves boundary delineation in the decoder stages. Across four MIS benchmarks, FLoMo-Net achieves superior boundary-aware performance and faster inference with fewer parameters than prior state-of-the-art 2D MIS models. Code is publicly available at https://github.com/rayhan-ahmed91/FLoMo-Net.

Teleportation in Virtual Reality (VR) is a locomotion technique that allows users to navigate between locations within a virtual environment instantly. Traditionally, VR teleportation is performed using physical controllers, where users control a teleportation pointer — represented by a straight line or parabola — and activate the teleportation to the target destination by pressing a button. Recent advances in hand and eye-tracking capabilities in Head-Mounted Displays (HMDs) enable designers to leverage hand and eye-based interactions to enhance the immersion and naturalness of controller-free VR usage. However, there has been limited research on comparing different controller-free methods for VR teleportation across various elevations. To address this gap, we conducted a user study exploring three controller-free pointing techniques (gaze, hand, and head), four confirmation modalities (finger pinch, eye-blink, dwell, and voice), and two types of teleportation pointers (linear and parabolic) for VR teleportation across various elevations. Our results show that head-based pointing was faster and more accurate than other techniques, with head and gaze achieving higher throughput than hand-based methods. For confirmation, finger pinch yielded the best performance in terms of task completion time and throughput, followed by dwell, voice, and eye-blink; dwell was the most accurate. The linear pointer outperformed the parabolic pointer in some contexts. Based on these findings, we propose design guidelines to enhance controller-free VR teleportation using various input modalities.

Surface defect segmentation (SDS) plays a crucial role in modern industrial inspection, driven by advancements in deep learning and computer vision. However, existing methods often struggle with defects that vary in size, shape, and appearance, especially when defects resemble the background. One key challenge is the effective fusion of low-level spatial and high-level semantic features, which conventional encoder-decoder networks handle inadequately, resulting in poor boundary precision and contextual inconsistency. To address these limitations, we propose MSAC-Net, a context-aware encoder-decoder architecture optimized for robust SDS. The encoder employs a dual-stream design: one stream captures semantic features via EfficientNetV2M, while the other enhances structural detail using a multi-scale convolutional cascade and channel-shuffle SEDNet module. A spatial attention-enhanced fusion module unifies these streams to improve feature integration. At the bottleneck, a spatially shifted MLP is combined with residual context features and a multi-scale aggregation module to capture long-range dependencies. The decoder incorporates cross-attention gates and ECA-enhanced residual convolution blocks to progressively refine segmentation. Evaluations on four benchmark datasets (DAGM2007, SD900, Magnetic Tile, and KolektorSDD2) show that MSAC-Net achieves an average mDSC of 95.15%, outperforming state-of-the-art methods due to superior multi-scale feature integration, attention-guided fusion, and a lightweight design, making it ideal for industrial deployment. • MSAC-Net is a multi-stream context aggregation network for segmenting defects with indistinct edges and variable shapes. • Employs channel shuffling, dilated convolutions, and attention to enhance defect segmentation with multi-contextual features. • Performs effective fusion of local contextual and global semantic features. • Outperforms state-of-the-art segmentation methods in four benchmark datasets. • Balances performance and computation, demonstrating MSAC-Net’s suitability for real-world defect inspection and localization.

Mobile devices have the potential to facilitate remote tasks through Augmented Reality (AR) solutions by integrating digital information into the real world. Although prior studies have explored Mobile Augmented Reality (MAR) for co-located collaboration, none have investigated the impact of various viewing attributes that can influence remote task performance, such as target object viewing angles, synchronization styles, or having a secondary small screen showing other users current view in the MAR environment. In this paper, we explore five techniques considering these attributes, specifically designed for two modes of remote tasks: collaborative and competitive. We conducted a user study employing various combinations of those attributes for both tasks. In both instances, results indicate users' optimal performance and preference for the technique that allows asynchronous viewing of object manipulations on the small screen. Overall, this paper contributes novel techniques for remote tasks in MAR, addressing aspects such as viewing angle and synchronization in object manipulation alongside secondary small-screen interfaces. Additionally, it presents the results of a user study evaluating the effectiveness, usability, and user preference of these techniques in remote settings and offers a set of recommendations for designing and implementing MAR solutions to enhance remote activities.

Deployed AI applications are often heavily con-strained in computational resources; to this end, a method of producing miniature, well-performing neural networks for 2D image segmentation is devised and applied to the Severstal Steel Defect Detection and Kidney Tumor Segmentation (KiTS19) Challenges. By limiting the width of U-Net and employing a full-domain activation function, a network of aggregated weak learners is able to achieve a mean F1score within 93% of the EfficientNetB0 baseline using only 0.5% of the trainable parameters. A similar network of aggregated strong learners matches the mean F1score of the baseline on the Severstal dataset using only 10% of the trainable parameters. Gradient boosting is then applied to the weak learners, achieving a mean F1score within 98% of the strong learner network on the Severstal dataset with approximately 20% of the FLOPS; the key insight is in constraining the O(n2) relationship between network width and FLOPS. The same approach is applied to the KiTS19 dataset with good success in kidney detection. Interestingly, the method does not perform as well on the much harder to isolate tumor class, and the authors explore some possible reasons. In analyzing the impact of the full-domain activation function, the authors show that density of information is promoted by significantly reduced peaks in layer outputs and a wider range of output values. The method has significant implications in constrained deployments, as many small devices could be used to compute the overall network.

Many active learning techniques have been used and described over the years, including team-based learning (TBL). While this technique is well established, it is only recently that analyses that compare it to other teaching techniques have been reported. In this paper, we evaluate the impact of team-based learning on two major concerns for computer science instructors: the drop/attrition rates, and students' success in CS1. The results show some major improvements both in terms of the drop rate and students' success, as measured by final exam grades. For example, the number of students obtaining 50% or more on the final exam has increased from 54% to 75.5%. Moreover, the drop rate has decreased from more than 30% to 6.4%.

First year computer science programming has always been a challenge for many students as the course expectation is not only for them to be able to understand programming concepts, but also to produce creative solutions to problems. Team-based learning seems a natural solution to increase the amount of practice each student will get, and to increase students' interest and confidence. The initial results of these two years of experimentation with team-based learning suggests that it helps reduce the dropping rate in the class to a reasonable level (10%) and give greater confidence to students in their ability to succeed. In this paper, we present how team-based learning has been adapted for our first semester programming class and we discuss the advantages of this techniques and difficulties encountered.