查看更多>>摘要：Registering preoperative 3D medical models to the corresponding regions of the individual in a reality scene a critical foundation for augmented reality-based (AR-based) surgical navigation systems. A challenge is finding an appropriate spatial mapping function from the medical coordinate system to the AR coordinate system. Our work focuses on registering 3D medical models to the intracranial structures using RGBD point clouds. The mapping function is calculated using local facial features and a scale-invariant coherent point drift (SI-CPD) algorithm that eliminates the scaling parameter. The local facial features significantly reduce mismatched features between the 3D medical models and the RGBD point clouds of the scene, while the proposed SI-CPD algorithm restricts the registration process to translation and rotation operations only. Results demonstrate that our method achieves a target registration error (TRE) of 1.2498 +/- 0.0829 mm on private medical datasets and superior registration accuracy on the public Stanford Bunny dataset. Compared to ICP-type methods, the SI-CPD algorithm demonstrates enhanced robustness in handling noise and outliers. Our work introduces novel methodology to automatically register 3D medical models to the head with high accuracy.
Abbass, Mohammed Y.Kasban, H.Elsharkawy, Zeinab F.
1.1-1.12页
查看更多>>摘要：Deep learning approaches have notable results in the area of computer vision applications. Our paper presents improved LYT-Net, a Lightweight YUV Transformer-based models, as an innovative method to improve low-light scenes. Unlike traditional Retinex-based methods, the proposed framework utilizes the chrominance (U and V) and luminance (Y) channels in YUV color-space, mitigating the complexity between color details and light in scenes. LYT-Net provides a thorough contextual realization of the image while keeping architecture burdens low. In order to tackle the issue of weak feature generation of traditional Channel-wise Denoiser (CWD) Block, improved CWD is proposed using Triplet Attention network. Triplet Attention network is exploited to capture both dynamics and static features. Qualitative and quantitative experiments demonstrate that the proposed technique effectively addresses images with varying exposure levels and outperforms state-of-the-art techniques. Furthermore, the proposed technique shows faster computational performance compared to other Retinex-based techniques, promoting it as a suitable option for real-time computer vision topics. The source code is available at https://github.com/Mohammed-Abbass/YUV-Attention
查看更多>>摘要：We present Bi-Scale density Plot (BSP), anew technique to enhance density plots by efficiently optimizing the local density variance in high- and mid-density regions while providing more details in low-density regions. When visualizing large and dense discrete point samples, scatterplots and thematic maps are often employed and we need density plots to further provide aggregated views. However, in the density plots, local patterns such as outliers can be filtered out and meaningful structures such as local density variations can be broken down. The key innovations in BSP include (i) the unified bin-summarize-decompose-combine framework for interactively bi-scale enhancing density plots through combining large- and small-scale density variations; and (ii) the variance-aware filter, which is reformulated based on the edge-preserving image filter, for maintaining the relative data density while reducing the excessive variability in the density plot. Further, BSP can be adopted with a 2D colormap, allowing simultaneous exploration of the enhanced structures and recovering the absolute aggregated densities to improve comparison and lookup tasks. We empirically evaluate our techniques in a controlled study and present two case studies to demonstrate their effectiveness in exploring large data.
查看更多>>摘要：Geometry quality assessment (GQA) of colorless point clouds is crucial for evaluating the performance of emerging point cloud-based solutions (e.g., watermarking, compression, and 3-Dimensional (3D) reconstruction). Unfortunately, existing objective GQA approaches are traditional full-reference metrics, whereas state-of-theart learning-based point cloud quality assessment (PCQA) methods target both color and geometry distortions, neither of which are qualified for the no-reference GQA task. In addition, the lack of large-scale GQA datasets with subjective scores, which are always imprecise, biased, and inconsistent, also hinders the development of learning-based GQA metrics. Driven by these limitations, this paper proposes a no-reference geometry-only quality assessment approach based on list-wise rank learning, termed LRL-GQA, which comprises of a geometry quality assessment network (GQANet) and a list-wise rank learning network (LRLNet). The proposed LRL-GQA formulates the no-reference GQA as a list-wise rank problem, with the objective of directly optimizing the entire quality ordering. Specifically, a large dataset containing a variety of geometry-only distortions is constructed first, named LRL dataset, in which each sample is label-free but coupled with quality ranking information. Then, the GQANet is designed to capture intrinsic multi-scale patch-wise geometric features in order to predict a quality index for each point cloud. After that, the LRLNet leverages the LRL dataset and a likelihood loss to train the GQANet and ranks the input list of degraded point clouds according to their distortion levels. In addition, the pre-trained GQANet can be fine-tuned further to obtain absolute quality scores. Experimental results demonstrate the superior performance of the proposed no-reference LRL-GQA method compared with existing full-reference GQA metrics. The source code can be found at: https://github.com/VCG-NJUST/LRL-GQA.
查看更多>>摘要：The layout of commercial space is crucial for enhancing user experience and creating business value. However, designing the layout of a mid-scale commercial space remains challenging due to the need to balance rationality, functionality, and safety. In this paper, we propose a novel method that utilizes the Centroidal Voronoi Tessellation (CVT) to generate commercial space layouts automatically. Our method is a multi-level spatial division framework, whereat each level, we create and optimize Voronoi diagrams to accommodate complex multi-scale boundaries. We achieve spatial division at different levels by combining the standard Voronoi diagrams with the rectangular Voronoi diagrams. Our method also leverages Voronoi diagrams' generation controllability and division diversity, offering customized control and diversity generation that previous methods struggled to provide. Extensive experiments and comparisons show that our method offers an automated and efficient solution for generating high-quality commercial space layouts.
查看更多>>摘要：Finding surface loops around narrow sections of a surface is widely used as a prepossessing step in various applications such as segmentation, shape analysis, path planning, and robotics. A common approach to locating such loops is based on surface topology. However, such geodesic loops also exist on topologically trivial genus-0 surfaces, where all such loops can continuously deform to a point. While a few existing 3D geometry-aware topological approaches may succeed in detecting such additional narrow loops, their construction can be cumbersome. To extend beyond the limitations of topologically nontrivial independent loops while remaining efficient, we propose a novel approach that leverages the shape's skeleton for computing surface loops of handle or tunnel type. Given a closed surface mesh, our algorithm produces a practically comprehensive set of loops encircling narrow regions of the volume inside or outside the surface. Notably, our approach streamlines and expedites computations by accepting a skeleton, a 1D representation of the shape, as part of the input. Specifically, handle-type loops are discovered by examining a small subset of the skeleton points as candidate loop centers, while tunnel-type loops are identified by examining only the high-valence skeleton points.
查看更多>>摘要：Inspired by Neural Radiance Field's (NeRF) groundbreaking success in novel view synthesis, current methods mostly employ variants of various deep neural network architectures, and use the combination of multi-scale feature maps with the target viewpoint to synthesize novel views. However, these methods only consider spatial domain features, inevitably leading to the loss of some details and edge information. To address these issues, this paper innovatively proposes the FreqSpace-NeRF (FS-NeRF), aiming to significantly enhance the rendering fidelity and generalization performance of NeRF in complex scenes by integrating the unique advantages of spectral domain and spatial domain deep neural networks, and combining contrastive learning driven data augmentation techniques. Specifically, the core contribution of this method lies in designing a dual-stream network architecture: on one hand, capturing global frequency features through Fourier transformation; on the other hand, finely refining local details using well-established spatial domain convolutional neural networks. Moreover, to ensure the model can more acutely distinguish subtle differences between different views, we propose two loss functions: Frequency-Space Contrastive Entropy Loss (FSCE Loss) and Adaptive Spectral Contrastive Loss (ASC Loss). This innovation aims to more effectively guide the data flow and focuses on minimizing the frequency discrepancies between different views. By comprehensively utilizing the fusion of spectral and spatial domain features along with contrastive learning, FS-NeRF achieves significant performance improvements in scene reconstruction tasks. Extensive qualitative and quantitative evaluations confirm that our method surpasses current state-of-the-art (SOTA) models in this field.
查看更多>>摘要：Low-light image enhancement (LLIE) aims to enhance the visibility and quality of low-light images. However, existing methods often struggle to effectively balance global and local image content, resulting in suboptimal results. To address this challenge, we propose a novel multi-scale wavelet feature fusion network (MWFFnet) for low-light image enhancement. Our approach utilizes a U-shaped architecture where traditional downsampling and upsampling operations are replaced by discrete wavelet transform (DWT) and inverse DWT (IDWT), respectively. This strategy helps to reduce the difficulty of learning the complex mapping from low-light images to well-exposed ones. Furthermore, we incorporate a dual transposed attention (DTA) module for each feature scale. DTA effectively captures long-range dependencies between image contents, thus enhancing the network's ability to understand intricate image structures. To further improve the enhancement quality, we develop a cross-layer attentional feature fusion (CAFF) module that effectively integrates features from both the encoder and decoder. This mechanism enables the network to leverage contextual information across various levels of representation, resulting in amore comprehensive understanding of the images. Extensive experiments demonstrate that with a reasonable model size, the proposed MWFFnet outperforms several state-of-the-art methods. Our code will be available online.2
查看更多>>摘要：Sports visualization analysis is an important area within visualization studies. However, there is a lack of tools tailored for NBA writers among existing systems. Creating these tools would improve understanding of the game's complex dynamics, particularly player interactions. We propose a visualization system to improve understanding of complex NBA game data. Featuring multiple modules, it allows users to analyze the game from various perspectives. This paper highlights the system's use of storylines to examine player interactions, enhancing the extraction of valuable insights. The study shows that our design enhances personalized in-game data analysis, improving the understanding and aiding in identifying critical moments.
查看更多>>摘要：Motorcycle graph is widely adopted as an intermediate block in state-of-art semi-structured quad meshing methods. However, constructing and simplifying it on 3D triangle meshes still face challenges in performance and stability. To address these challenges, we present a novel motorcycle graph construction and simplification method for semi-structured quad mesh generation. First, we introduce a piecewise advancing algorithm on parameterized triangle meshes with specially designed data structures to ensure reliable and high-performing motorcycle graph tracing. Second, we enhance the existing zero-collapse procedure with non-intersecting paths creation and feature preserving for T-mesh simplification. Third, we integrate our motorcycle graph construction and simplification algorithm into the state-of-art semi-structured quad meshing pipeline. A comparison with typical state-of-art methods proves that our method can generate quad meshes with superior topological quality and feature preservation capability. We also conduct batch experiments to demonstrate the efficiency, robustness of the proposed method.
NETL
NSTL
Elsevier