We propose a novel framework for Russian Roulette and Splitting (RRS) tailored to wavefront path tracing, a highly parallel rendering architecture that processes path states in batched, stage-wise execution for efficient GPU utilization. Traditional RRS methods, with unpredictable path counts, are fundamentally incompatible with wavefront's preallocated memory and scheduling requirements. To resolve this, we introduce a normalized RRS formulation with a bounded path count, enabling stable and memory-efficient execution. Furthermore, we pioneer the use of neural networks to learn RRS factors, presenting two models: NRRS and AID-NRRS. At a high level, both feature a carefully designed RRSNet that explicitly incorporates RRS normalization, with only subtle differences in their implementation. To balance computational cost and inference accuracy, we introduce Mix-Depth, a path-depth-aware mechanism that adaptively regulates neural evaluation, further improving efficiency. Extensive experiments demonstrate that our method outperforms traditional heuristics and recent RRS techniques in both rendering quality and performance across a variety of complex scenes.

3DGS is an emerging and increasingly popular technology in the field of novel view synthesis. Its highly realistic rendering quality and real-time rendering capabilities make it promising for various applications. However, when applied to large-scale aerial urban scenes, 3DGS methods suffer from issues such as excessive memory consumption, slow training times, prolonged partitioning processes, and significant degradation in rendering quality due to the increased data volume. To tackle these challenges, we introduce \textbf{HUG}, a novel approach that enhances data partitioning and reconstruction quality by leveraging a hierarchical neural Gaussian representation. We first propose a visibility-based data partitioning method that is simple yet highly efficient, significantly outperforming existing methods in speed. Then, we introduce a novel hierarchical weighted training approach, combined with other optimization strategies, to substantially improve reconstruction quality. Our method achieves state-of-the-art results on one synthetic dataset and four real-world datasets.

Novel view synthesis for dynamic scenes presents a significant challenge in computer graphics. While recent 3D Gaussian splatting methods have achieved state-of-the-art quality and speed for static scenes, their direct extension to 4D dynamic scenes remains non-trivial. Existing approaches for dynamic scene novel view synthesis typically fall into two categories: either employing time-varying dynamic Gaussians, which often suffer from artifacts due to MLP limitations, or directly extending Gaussians to 4D, which, despite yielding high rendering quality, incurs substantial memory overhead. This paper introduces a novel 4D anchor-based framework that effectively leverages the stronger representational power of 4D Gaussians while crucially addressing their storage inefficiency. Our approach models dynamic scenes by binding Gaussians to strategically distributed anchor points. Furthermore, we propose a dynamic anchor control strategy to generate additional anchors in dynamic regions requiring detailed reconstruction. Additionally, we design an anchor stabilization strategy to fix the attributes of anchors in static regions during training, thereby preventing redundancy. Extensive experiments on various benchmarks, including N3DV and the Technicolor dataset, demonstrate the superior visual quality of our method.

Recent research on learnable neural representations has been widely adopted in the field of 3D scene reconstruction and neural rendering applications. However, traditional feature grid representations often suffer from a substantial memory footprint, posing a significant bottleneck for modern parallel computing hardware. In this paper, we present neural vertex features, a generalized formulation of learnable representation for neural rendering tasks involving explicit mesh surfaces. Instead of uniformly distributing neural features throughout 3D space, our method stores learnable features directly at mesh vertices, leveraging the underlying geometry as a compact and structured representation for neural processing. This not only optimizes memory efficiency, but also improves feature representation by aligning compactly with the surface using task-specific geometric priors. Additionally, neural vertex features offer improved feature representation by compactly aligning with the surface using task-specific geometric priors. We validate our neural representation across diverse neural rendering tasks, with a specific emphasis on neural radiosity. Experimental results demonstrate that our method reduces memory consumption to only one-fifth (or even less) of grid-based representations, while maintaining comparable rendering quality and lowering inference overhead.

Recently, deep learning based multi-view stereo (MVS) networks have demonstrated their excellent performance on various benchmarks. In this paper, we present an effective and efficient recurrent neural network (RNN) for accurate and complete dense point cloud reconstruction. Instead of regularizing the cost volume via conventional 3D CNN or unidirectional RNN like previous attempts, we adopt a bidirectional hybrid Long Short-Term Memory (LSTM) based structure for cost volume regularization. The proposed bidirectional recurrent regularization is able to perceive full-space context information comparable to 3D CNNs while saving runtime memory. For post-processing, we introduce a visibility based approach for depth map refinement to obtain more accurate dense point clouds. Extensive experiments on DTU, Tanks and Temples and ETH3D datasets demonstrate that our method outperforms previous state-of-the-art MVS methods and exhibits high memory efficiency at runtime.

Image interpolation has a wide range of applications such as frame rate-up conversion and free viewpoint TV. Despite significant progresses, it remains an open challenge especially for image pairs with large displacements. In this paper, we first propose a novel optimization algorithm for motion estimation, which combines the advantages of both global optimization and a local parametric transformation model. We perform optimization over dynamic label sets, which are modified after each iteration using the prior of piecewise consistency to avoid local minima. Then we apply it to an image interpolation framework including occlusion handling and intermediate image interpolation. We validate the performance of our algorithm experimentally, and show that our approach achieves state-of-the-art performance.

With promising benefits such as emission reduction, traffic congestion alleviation and parking space saving, electric vehicle sharing systems have attracted increasing attentions. This paper proposes a Trip Pricing Scheme (TPS) for a large-scale EV-sharing network with Shared Electric Vehicle (SEV) demand prediction. In the proposed system, the SEV traffic demand is firstly predicted through a model which includes a cascade graph convolutional neural network and a long-short term memory neural network. Based on this, the TPS is modelled as a mixed-integer nonlinear programming problem, which aims at maximizing the total system profit from EV sharing business. The proposed TPS determines the optimal combination of the two price adjustment levels, which provides incentives to the spatial-temporal distribution of SEVs’ traffic flows to maximize the system’s profit. Numerical simulations are conducted to validate the effectiveness of the proposed method.

In this paper, we present a framework of view synthesis, including range guided depth refinement and uncertainty-aware aggregation based novel view synthesis. We first propose a novel depth refinement method to improve the quality and robustness of the depth map reconstruction. To that end, we use a range prior to constrain the estimated depth, which helps us to get more accurate depth information. Then we propose an uncertainty-aware aggregation method for novel view synthesis. We compute the uncertainty of the estimated depth for each pixel, and reduce the influence of pixels whose uncertainty are large when synthesizing novel views. This step helps to reduce some artifacts such as ghost and blur. We validate the performance of our algorithm experimentally, and we show that our approach achieves state-of-the-art performance.

In this paper, we present a novel recurrent multi-view stereo network based on long short-term memory (LSTM) with adaptive aggregation, namely AA-RMVSNet. We firstly introduce an intra-view aggregation module to adaptively extract image features by using context-aware convolution and multi-scale aggregation, which efficiently improves the performance on challenging regions, such as thin objects and large low-textured surfaces. To overcome the difficulty of varying occlusion in complex scenes, we propose an interview cost volume aggregation module for adaptive pixel-wise view aggregation, which is able to preserve better-matched pairs among all views. The two proposed adaptive aggregation modules are lightweight, effective and complementary regarding improving the accuracy and completeness of 3D reconstruction. Instead of conventional 3D CNNs, we utilize a hybrid network with recurrent structure for cost volume regularization, which allows high-resolution reconstruction and finer hypothetical plane sweep. The proposed network is trained end-to-end and achieves excellent performance on various datasets. It ranks 1stamong all submissions on Tanks and Temples benchmark and achieves competitive results on DTU dataset, which exhibits strong generalizability and robustness. Implementation of our method is available at https://github.com/QT-Zhu/AA-RMVSNet.

Wide-baseline image interpolation is useful in many multimedia applications such as virtual street roaming and 3D TV. It is also a challenging problem because the large translations and rotations of image patches make it hard to estimate the motion fields between wide-baseline image pairs. We propose a refinement strategy based on salient error detection to improve the result of existing approaches of wide-baseline image interpolation, where we combine the advantages of methods based on piecewise-linear transformation and methods based on variational model. We first use a lightweight interpolation method to estimate the initial motion field between the input image pair, and synthesize the intermediate image as the initial result. Then we detect regions with noticeable artifacts in the initial image to find areas whose motion vectors should be refined. Finally, we refine the motion field of the detected regions using a variational model based method, and obtain the refined intermediate image. The refinement strategy of our method can be used as the post refinement step for many other image interpolation algorithms. We show the effectiveness and efficiency of our method through experiments on different datasets.

Misclassification in semantic segmentation mostly occurs in the pixels around the semantic contour. In this work, we address the task of aerial image segmentation by borrowing the kernel prior from classical edge detecting operator. We propose a module called Sobel Heuristic Kernel(SHK). Our work makes several main contributions and experimentally shows good performance. To the best of our knowledge, we are the first to combine traditional edge detection method and deep learning method in semantic segmentation. Our SHK module reaches state of the art in the Inria Aerial Image Labeling dataset.

In this paper, we introduce a method to augment virtual environment with multiple videos. Our goal is to make a virtual-real fusion system that fuses dynamic imagery with 3D models in a real-time display, so as to help observers visualize dynamic videos simultaneously in the context of 3D models. 3D models in virtual environment are reconstructed using multiple view vision methods. Based on this, images can be registered through feature matching among images. Foreground objects in videos lead to distortions when video images are simply projected to static models. Detection and tracking of those objects are needed, and then several ordinary 3D models are used to represent those objects. Both geometry and appearance are taken into account to recover characteristic of different objects. This paper focuses on the integration of these components into a prototype system and the presentation of results shows the benefits of an virtual-real fusion system.

We present an adaptive figure-ground segmentation algorithm that is capable of extracting foreground objects in a generic environment. Starting from an interactively assigned background mask, an initial background prior is defined and multiple soft-label partitions are generated from different foreground priors by progressive patch merging. These partitions are fused to produce a foreground probability map. The probability map is then binarized via threshold sweeping to create multiple hard-label candidates. A set of segmentation hypotheses is formed using different evaluation scores. From this set, the hypothesis with maximal local stability is propagated as the new background prior, and the segmentation process is repeated until convergence. Similarity voting is used to select a winner set, and the corresponding hypotheses are fused to yield the final segmentation result. Experiments indicate that our method performs at or above the current state-of-the-art on several data sets, with particular success on challenging scenes that contain irregular or multiple-connected foregrounds.

This paper proposes a 3D-2D registration method to determine the pose of a camera from an uncalibrated image of a given 3D model. Through the combined use of global silhouette comparison and local feature point matching, the camera pose of a 2D photograph can be accurately determined by minimising a hybird objective function. The registered photos can be used to stitch textures or refine the geometry of 3D models. This method can be used independently or collaboratively in image-based modeling systems to effectively treat restrictive camera-model position configurations. The experiment shows that our method is more flexible than the methods using only geometrical or local feature information.

We propose an adaptive figure-ground classification algorithm to automatically extract a foreground region using a user-provided bounding-box. The image is first over-segmented with an adaptive mean-shift algorithm, from which background and foreground priors are estimated. The remaining patches are iteratively assigned based on their distances to the priors, with the foreground prior being updated online. A large set of candidate segmentations are obtained by changing the initial foreground prior. The best candidate is determined by a score function that evaluates the segmentation quality. Rather than using a single distance function or score function, we generate multiple hypothesis segmentations from different combinations of distance measures and score functions. The final segmentation is then automatically obtained with a voting or weighted combination scheme from the multiple hypotheses. Experiments indicate that our method performs at or above the current state-of-the-art on several datasets, with particular success on challenging scenes that contain irregular or multiple-connected foregrounds. In addition, this improvement in accuracy is achieved with low computational cost.

This paper proposes an algorithm that solves planar homography by iterative linear optimization. We iteratively employ direct linear transformation (DLT) algorithm to robustly estimate the homography induced by a given set of point correspondences under perspective transformation. By simple on-the-fly homogeneous coordinate adjustment we progressively minimize the difference between the algebraic error and the geometric error. When the difference is sufficiently close to zero, the geometric error is equivalently minimized and the homography is reliably solved. Backward covariance propagation is employed to do error analysis. The experiments prove that the algorithm is able to find global minimum despite erroneous initialization. It gives very precise estimate at low computational cost and greatly outperforms existing techniques.

Streamline based method is one of the most important vector field visualization methods. In the past streamline placements algorithms, little physical related feature was considered, which is very important in our opinion. A novel streamline placement algorithm for 3D vector field is introduced in this paper. We measure the difference between the inflow and the outflow to evaluate the local spatial-varying feature at a specified field point. A Difference of Inflow and Outflow Matrix (DIOM) is then calculated to describe the global appearance of the field. We draw streamlines by choosing the local extreme points in DIOM as seeds. DIOM is somewhat like flow divergence and is physics-related thus re-flects intrinsic characteristics of the vector field. The strategy performs well in revealing features of the vector field even with relatively few streamlines both in 3D vector field

Abstract: A new classification algorithm based on multivariate analysis is proposed to discover and simulate the grading policy on school transcript data sets. The framework comprises three major steps. First, factor analysis is adopted to separate the scores of several different subjects into grading‐related ones and grading‐unrelated ones. Second, multidimensional scaling is employed for dimensionality reduction to facilitate subsequent data visualization and interpretation. Finally, a support vector machine is trained to classify the filtered data into different grades. This work provides an attractive framework for intelligent data analysis and decision making. It also exhibits the advantages of high classification accuracy and supports intuitive data interpretation.

A new knowledge mining framework based on multivariate analyses is proposed to discover and simulate the school grading policy. The framework comprises three major steps. Firstly, factor analysis is adopted to separate the scores of several different subjects into grading-related ones and grading-unrelated ones. Secondly, multidimensional scaling is employed for dimensionality reduction to facilitate subsequent data visualization and interpretation. Finally, a support vector machine is trained to classify the filtered data into different grades. This work provides an attractive framework for intelligent data analysis and decision-making. It also exhibits the advantages of high classification accuracy and supports intuitive data interpretation.

A quick and reliable model-based head motion tracking scheme is presented. In this approach, rigid head motion and non-rigid facial animation are robustly tracked simultaneously by statistically analyzing the local regions of several representative facial features. The features are defined and operated based on a mesh model that helps maintain a global constraint on the local features and avoid the time-consuming appearance computation. A statistical model is computed from a moderate training set that is obtained by synthesizing different poses from a given standard initial image. During tracking, feature-based local distributions are obtained directly from the video frames and the troublesome feature detection or model rendering process is avoided. The observed distribution is compared with the pre-computed statistical model and the tracking is achieved by minimizing an error function based on the maximum likelihood principle. Experimental results show that this tracking strategy is robust to a wide range of head motion, facial animation and partial occlusion. The tracking can be conducted in nearly real-time and is easy to recover from failures.

An efficient model-independent 3D texture synthesis algorithm based on texture growing and texture turbulence is presented to simulate vivid 3D solid textures from 2D growable texture patterns. Given a 2D texture pattern of some growable material, our algorithm is able to create a tileable anisotropic 3D texture pattern to simulate the natural property of the material. Target objects are directly dipped into the 3D texture pattern to generate creative, sculpture like models that can be presented with reasonable interactive frame rates. Additionally, our method is conceptually simple, computationally fast, and storage efficient. To the best of our knowledge, this is the first approach that transfers a given 2D texture naturally to point rendering systems

Support Vector Machine (SVM) is a new learning method developed in recent years based on the foundations of statistical learning theory. By taking a transductive approach instead of an inductive one in support vector classifiers, the test set can be used as an additional source of information about margins. Intuitively, we would expect transductive learning to yield improvements when the training sets are small or when there is a significant deviation between the training and working set subsamples of the total population. In this paper, a progressive transductive support vector machine is addressed to extend Joachims' Transductive SVM to handle different class distributions. It solves the problem of having to estimate the ratio of positive/negative examples from the working set. The experimental results show that the algorithm is very promising.