We present ScaleFree, a GPU-accelerated adaptive Kernel Density Estimation (KDE) algorithm for scalable, interactive multiscale point cloud exploration. With this technique, we cater to the massive datasets and complex multiscale structures in advanced scientific computing, such as cosmological simulations with billions of particles. Effective exploration of such data requires a full 3D understanding of spatial structures, a capability for which immersive environments such as VR are particularly well suited. However, simultaneously supporting global multiscale context and fine-grained local detail remains a significant challenge. A key difficulty lies in dynamically generating continuous density fields from point clouds to facilitate the seamless scale transitions: while KDE is widely used, precomputed fields restrict the accuracy of interaction and omit fine-scale structures, while dynamic computation is often too costly for real-time VR interaction. We address this challenge by leveraging GPU acceleration with k-d-tree-based spatial queries and parallel reduction within a thread group for on-the-fly density estimation. With this approach, we can recalculate scalar fields dynamically as users shift their focus across scales. We demonstrate the benefits of adaptive density estimation through two data exploration tasks: adaptive selection and progressive navigation. Through performance experiments, we demonstrate that ScaleFree with GPU-parallel implementation achieves orders-of-magnitude speedups over sequential and multi-core CPU baselines. In a controlled experiment, we further confirm that our adaptive selection technique improves accuracy and efficiency in multiscale selection tasks.

Metal defect semantic segmentation is a crucial process for classifying and locating defects during the industrial production process, which holds paramount importance in elevating the quality of metal products. Recently, deep learning has exhibited impressive capabilities in identifying and segmenting defects on metal surfaces . However, the prevalent use of fully supervised segmentation techniques demands a substantial amount of annotated data for effective model training, which is hard to obtain in real scenarios. Additionally, most defects of metal products exhibit indistinct edge details, which hinders precise defect localization. In this study, a Crop-Paste and diffusion-based semi-supervised segmentation network (CPDNet) is proposed to identify pixel-level defects on metal surfaces by utilizing data that are both labeled and unlabeled. Firstly, a semi-supervised training method Crop-Paste is proposed to facilitate the learning of comprehensive semantic features from an extensive of unlabeled images and a restricted set of labeled images. Secondly, a frequency-directed diffusion model is proposed to recover high frequency features of defects to generate more accurate segmentation results. Lastly, an edge aware module is proposed in Sobel mean-teacher (M-T) UNet to improve the boundary information representation associated with defects. The experimental results on four datasets related to metal surface defects and a multimodal dataset show that CPDNet achieves a better performance in comparison with those state-of-the-art methods.

We propose and study a novel cross-reality environment that seamlessly integrates a monoscopic 2D surface (an interactive screen with touch and pen input) with a stereoscopic 3D space (an augmented reality HMD) to jointly host spatial data visualizations. This innovative approach combines the best of two conventional methods of displaying and manipulating spatial 3D data, enabling users to fluidly explore diverse visual forms using tailored interaction techniques. Providing such effective 3D data exploration techniques is pivotal for conveying its intricate spatial structures-often at multiple spatial or semantic scales-across various application domains and requiring diverse visual representations for effective visualization. To understand user reactions to our new environment, we began with an elicitation user study, in which we captured their responses and interactions. We observed that users adapted their interaction approaches based on perceived visual representations, with natural transitions in spatial awareness and actions while navigating across the physical surface. Our findings then informed the development of a design space for spatial data exploration in cross-reality. We thus developed cross-reality environments tailored to three distinct domains: for 3D molecular structure data, for 3D point cloud data, and for 3D anatomical data. In particular, we designed interaction techniques that account for the inherent features of interactions in both spaces, facilitating various forms of interaction, including mid-air gestures, touch interactions, pen interactions, and combinations thereof, to enhance the users' sense of presence and engagement. We assessed the usability of our environment with biologists, focusing on its use for domain research. In addition, we evaluated our interaction transition designs with virtual and mixed-reality experts to gather further insights. As a result, we provide our design suggestions for the cross-reality environment, emphasizing the interaction with diverse visual representations and seamless interaction transitions between 2D and 3D spaces.

We propose three novel spatial data selection techniques for particle data in VR visualization environments. They are designed to be target- and context-aware and be suitable for a wide range of data features and complex scenarios. Each technique is designed to be adjusted to particular selection intents: the selection of consecutive dense regions, the selection of filament-like structures, and the selection of clusters-with all of them facilitating post-selection threshold adjustment. These techniques allow users to precisely select those regions of space for further exploration-with simple and approximate 3D pointing, brushing, or drawing input-using flexible point- or path-based input and without being limited by 3D occlusions, non-homogeneous feature density, or complex data shapes. These new techniques are evaluated in a controlled experiment and compared with the Baseline method, a region-based 3D painting selection. Our results indicate that our techniques are effective in handling a wide range of scenarios and allow users to select data based on their comprehension of crucial features. Furthermore, we analyze the attributes, requirements, and strategies of our spatial selection methods and compare them with existing state-of-the-art selection methods to handle diverse data features and situations. Based on this analysis we provide guidelines for choosing the most suitable 3D spatial selection techniques based on the interaction environment, the given data characteristics, or the need for interactive post-selection threshold adjustment.