In this paper, we propose Scene Splatter, a momentum-based paradigm for video diffusion to generate generic scenes from single image. Existing methods, which employ video generation models to synthesize novel views, suffer from limited video length and scene inconsistency, leading to artifacts and distortions during further reconstruction. To address this issue, we construct noisy samples from original features as momentum to enhance video details and maintain scene consistency. However, for latent features with the perception field that spans both known and unknown regions, such latent-level momentum restricts the generative ability of video diffusion in unknown regions. Therefore, we further introduce the aforementioned consistent video as a pixel-level momentum to a directly generated video without momentum for better recovery of unseen regions. Our cascaded momentum enables video diffusion models to generate both high-fidelity and consistent novel views. We further finetune the global Gaussian representations with enhanced frames and render new frames for momentum update in the next step. In this manner, we can iteratively recover a 3D scene, avoiding the limitation of video length. Extensive experiments demonstrate the generalization capability and superior performance of our method in high-fidelity and consistent scene generation.

Recent studies on Vision-Language-Action (VLA) models have shifted from the end-to-end action-generation paradigm toward a pipeline involving task planning followed by action generation, demonstrating improved performance on various complex, long-horizon manipulation tasks. However, existing approaches vary significantly in terms of network architectures, planning paradigms, representations, and training data sources, making it challenging for researchers to identify the precise sources of performance gains and determine which component is more difficult to learn. To systematically investigate the impacts of different planning paradigms and representations isolating from network architectures and training data, in this paper, we introduce \name, a unified VLA architecture suite capable of various task planning paradigms, and design a comprehensive suite of controlled experiments across diverse object categories (rigid and deformable), visual modalities (2D and 3D), environments (simulation and real-world), and end-effectors (grippers and dexterous hands). Our results demonstrate that: 1) visually grounded planning representations are generally better than language planning representations; 2) the Hierarchical-VLA paradigm generally achieves superior performance than other paradigms, albeit at the cost of slower training and inference speeds.

Point clouds captured by different sensors such as RGB-D cameras and LiDAR possess non-negligible domain gaps. Most existing methods design different network architectures and train separately on point clouds from various sensors. Typically, point-based methods achieve outstanding performances on even-distributed dense point clouds from RGB-D cameras, while voxel-based methods are more efficient for large-range sparse LiDAR point clouds. In this paper, we propose geometry-to-voxel auxiliary learning to enable voxel representations to access point-level geometric information, which supports better generalisation of the voxel-based backbone with additional interpretations of multi-sensor point clouds. Specifically, we construct hierarchical geometry pools generated by a voxel-guided dynamic point network, which efficiently provide auxiliary fine-grained geometric information adapted to different stages of voxel features. We conduct experiments on joint multi-sensor datasets to demonstrate the effectiveness of GeoAuxNet. Enjoying elaborate geometric information, our method outperforms other models collectively trained on multi-sensor datasets, and achieve competitive results with the-state-of-art experts on each single dataset.

3D Gaussian Splatting (3DGS) has demonstrated impressive novel view synthesis performance. While conventional methods require per-scene optimization, more recently several feed-forward methods have been proposed to generate pixel-aligned Gaussian representations with a learnable network, which are generalizable to different scenes. However, these methods simply combine pixel-aligned Gaussians from multiple views as scene representations, thereby leading to artifacts and extra memory cost without fully capturing the relations of Gaussians from different images. In this paper, we propose Gaussian Graph Network (GGN) to generate efficient and generalizable Gaussian representations. Specifically, we construct Gaussian Graphs to model the relations of Gaussian groups from different views. To support message passing at Gaussian level, we reformulate the basic graph operations over Gaussian representations, enabling each Gaussian to benefit from its connected Gaussian groups with Gaussian feature fusion. Furthermore, we design a Gaussian pooling layer to aggregate various Gaussian groups for efficient representations. We conduct experiments on the large-scale RealEstate10K and ACID datasets to demonstrate the efficiency and generalization of our method. Compared to the state-of-the-art methods, our model uses fewer Gaussians and achieves better image quality with higher rendering speed.

Abstract The Coronavirus Disease 2019 (COVID‐19) epidemic has constituted a Public Health Emergency of International Concern. Chest computed tomography (CT) can help early reveal abnormalities indicative of lung disease. Thus, accurate and automatic localisation of lung lesions is particularly important to assist physicians in rapid diagnosis of COVID‐19 patients. The authors propose a classifier‐augmented generative adversarial network framework for weakly supervised COVID‐19 lung lesion localisation. It consists of an abnormality map generator, discriminator and classifier. The generator aims to produce the abnormality feature map M to locate lesion regions and then constructs images of the pseudo‐healthy subjects by adding M to the input patient images. Besides constraining the generated images of healthy subjects with real distribution by the discriminator, a pre‐trained classifier is introduced to enhance the generated images of healthy subjects to possess similar feature representations with real healthy people in terms of high‐level semantic features. Moreover, an attention gate is employed in the generator to reduce the noise effect in the irrelevant regions of M . Experimental results on the COVID‐19 CT dataset show that the method is effective in capturing more lesion areas and generating less noise in unrelated areas, and it has significant advantages in terms of quantitative and qualitative results over existing methods.

Many real-world optimization problems involve uncertainties, and in such situations it is often desirable to identify robust solutions that perform well over the possible future scenarios. In this paper, we focus on input uncertainty, such as in manufacturing, where the actual manufactured product may differ from the specified design but should still function well. Estimating a solution's expected fitness in such a case is challenging, especially if the fitness function is expensive to evaluate, and its analytic form is unknown. One option is to average over a number of scenarios, but this is computationally expensive. The archive sample approximation method reduces the required number of fitness evaluations by reusing previous evaluations stored in an archive. The main challenge in the application of this method lies in determining the locations of additional samples drawn in each generation to enrich the information in the archive and reduce the estimation error. In this paper, we use the Wasserstein distance metric to approximate the possible benefit of a potential sample location on the estimation error, and propose new sampling strategies based on this metric. Contrary to previous studies, we consider a sample's contribution for the entire population, rather than inspecting each individual separately. This also allows us to dynamically adjust the number of samples to be collected in each generation. An empirical comparison with several previously proposed archive-based sample approximation methods demonstrates the superiority of our approaches.

The fast and frequent movement of vehicles creates many challenges in vehicular networks, such as handling regular topological changes. Predicting a vehicle's future location by preemptively adjusting to changes caused by vehicle movements is a potential solution to many of these problems. However, reliably predicting vehicle movement remains an issue due to its stochastic nature. This paper proposes a prediction method that probabilistically analyzes the vehicle's current movement to determine the vehicle's future steps. This is accomplished by combining the Kalman filter and hidden Markov model to include both temporal and historical data, thus improving the prediction reliability by the consideration of more system variables. The proposed approach is tested and compared to other recent approaches through simulation using SUMO and NS-2. The results show a 50% prediction error reduction of the proposed approach in comparison to other methods.

Vehicular networks are a rapidly growing technology with vehicle manufacturers already taking steps for its implementation. However, adding IP services to vehicular networks remains an issue. This is in part due to the mobile IP handover causing overhead and latencies unsuitable for the faster movement of vehicles. Frequent topological changes caused by this movement require smooth IP handovers due to the regularity in which they occur. In this paper, we propose a predictive handover utilizing a movement prediction method to conduct the costly handover ahead of time, thus providing a consistent IP connection. The approach is tested and compared in a variety of simulated vehicle network environments. Results show an overall improvement in network performance with handover latency greatly reduced.

The RSU deployment algorithms to cover interest regions have emerged as promising areas of research in vehicular ad hoc networks. However, there is few research on the discovery of the interest regions. Besides, the deployment algorithms are usually proposed for either continuous coverage or sparse coverage. The wisdom of the model selection are rarely addressed. In this paper, we proposed a new algorithm to discover interest regions. A budget-constrained coverage selection algorithm is also presented to help network designers choose suitable coverage models to meet the budget and quality requirements. The algorithms are implemented on top of Ns2 and the simulations are carried out using SUMO and OpenStreet Maps. The performance comparison between our algorithm and other two clustering algorithms prove that our algorithm has a better performance in terms of contact time for both sparse coverage and continue coverage.

Time-sensitive mobile service with vehicular networks depend on continuous coverage on the road system. It is a challenge to meet a tight budget while maintaining coverage quality. In order to resolve the problem, we model the road system as an undirected graph and reduce it to a Knapsack Constrained Steiner Tree model. Since the reduced model is an NP-hard problem, we resolve it by using Lagrangian decomposition approach. The terminal nodes in the graph are hotspots, where most vehicles accumulate; and, the paths connecting these hotspots are measured by a new metric: coverage value. We evaluated the our scheme by comparing it with the Maximum Continuous Coverage protocol in terms of the packet drop rate in the NS2 simulation. The final result shows that our coverage is reliable, and suitable for urban vehicular networks.

In a vehicular ad hoc network, vehicles can communicate with their correspondent nodes through the Internet using various wireless technologies. This kind of wireless networks brings a lot of new applications to metropolitan networks, such as online games, VoIP, etc. However, compared with traditional wireless networks, the high speed of vehicles and the limited transmission range of antennas introduce more frequent handoffs. During the handoff, vehicles have to switch their access points to establish new connections and the current wireless communications will be ceased temporarily. Therefore, long handoff latency causes poor throughput and obvious jitter. In this paper, a fast handoff scheme is proposed for vehicular networks. Using this scheme, vehicles are divided into different clusters; and within each cluster, the network mobility solution is used to reduce the total number of handoffs. Moreover, before the mobile routers start actual handoff process, they can receive their new care of addresses through assistant nodes in the same cluster. Simulation results demonstrate that handoff latency can be significantly reduced by our scheme.

Improving the performance of data gathering while saving the energy of network is one of the subjects of mobile data gathering protocols. In order to achieve the goal, the virtual force technique is introduced into the research where the data in the transmission queue and the distance between the sensor and the collector are simulated as forces. The collectors move on the direction of net force until the system reaches an equilibrium state. However, in the virtual force based data gathering system, the connectivity and communication quality have to be maintained for updating the changes of virtual forces. Otherwise, the loss of message will lead the system into a wrong equilibrium state and lower the performance of data gathering. In this paper, we formulate the moving process of data collectors as a Markov chain and determine the moving path using Markov decision process (MDP). The collectors move along the path defined by the optimized policy which is off-line computed and downloaded to collectors. Because the movement of collector no more relies on the online computed virtual force, the connectivity and communication quality of network have less effect on our MDP based protocol than the original virtual force based protocol. We compare the simulation results of our protocol, virtual force based protocol and the travel salesman problem (TSP) based protocol. The simulation results show that our protocol is resistant to connectivity error and works well when the communication error is applied to the system.

Nowadays wireless sensor network have been used in many area for automatic localization and target tracking. However, the large resource demands of sensors for simultaneous running might not be feasible and will lead to loss of functions. In this paper, we consider the problem of dynamical sensor selection based on user-defined objectives, such as maximizing k-coverage and detection probability. We model the problem as a POMDP, which selects k sensors in each time frame to maintain the area coverage quality. In order to fulfill the users chosen objective, we change the system behavior by changing the reward function of POMDP and the parameter of coverage model. The performance of our scheme is compared to three different schemes: RandomSelection, ShortestDistance and POMDP with Boolean model. The POMDP scheme with probability coverage model shows the best performance in terms of coverage and detection probability.

In order to optimize the performance of wireless sensor networks, based on the boolean sensing model (BSM) many coverage preserving protocols have been proposed; however, BSM model is too idealistic to be maintained in many applications. Sensing range could be changed by environment factors such as the obstacle and weather condition. BSM based protocols can not achieve their design goals because of the errors of coverage measurement. In this paper, we proposed our irregular sensing range detection model (RDM) which solves the range detection problem through estimating the sensing range by a revised ¿-shape algorithm. A RDM based distributed algorithm for measuring the area coverage is proposed. Its complexity are analyzed and compared to the centralized ¿-shape based area coverage algorithm. A set of simulation experiments is carried out to evaluate the performance of our algorithm in terms of coverage accuracy, coverage error and complexity. The result shows that the boolean sensing model will lead to errors when obstacles are injected into environment; however, our range detection model can detect the changes of environments and help protocols adapt to it.

Wireless video sensor networks (WVSN) can retrieve video/images from environment of interest, which can greatly enhance our knowledge of the physical world and pave the way for new applications. Since video signal is sensitive to temporal differences between sensor nodes and sink nodes, time synchronization is a critical service for WVSN. However, the existing wireless sensor network (WSN) time synchronization protocols cannot fulfill WVSN's new synchronization requirements. In this report, a new time synchronization technique, V-square, is proposed. Compared with existing WSN time synchronization protocols, V-square can reduce error in estimations on local clock parameters with very low communication and computation overhead. The performance of V-square is examined through both analysis and simulations.

Coverage is one of the fundamental measurements of quality in wireless sensor networks. In order to prolong the network lifetime while maintaining coverage, many node partition algorithms have been developed. In this article, we model coverage problem by a set coverage problem. Based on the density of information, the optimal node partitions are investigated by solving an ILP problem. An intersection point method (IPM) is introduced to reduce the number of variables in ILP to O(km) where m is the number of deployed sensors; k is the number of neighbors. Even though the ILP model can give an approximately optimal solution for generating a minimum cover set. It cannot be used in a distributed scenario. Based on the Voronoi Diagram we present a distributed partition algorithm that constructs minimum node partitions by merging voronoi cells. The simulation results show that the IPM based ILP coverage model can deal with extremely large areas and improve the ILP performance by reducing the number of variables and constraints. The voronoi based distributed partition algorithm can give approximate optimal results as given by the IPM based ILP solution. Both the flexibility and accuracy of our algorithms show the potential to be used in scheduling and duty circle algorithms.

Recent years, coverage has been widely investigated as one of the fundamental quality measurements of wireless sensor networks. In order to maintaining the coverage while saving energy of networks, algorithms have been developed to keep a minimum cover set of sensors working and turn off the redundant sensors. Generally, centralized algorithms can give a better result than distributed algorithms in terms of the number of active sensors. However, the heavy computation requirements and message overhead for collecting geographical location data keep centralized algorithms out of most distributed scenarios. In this article, Based on the idea of coverage compensation a distributed node partition algorithm for random deployments is presented to generate a minimum cover set by using the optimal node distributions created by the centralized algorithms such as GA. A Genetic Algorithm for coverage is proposed too to demonstrate how an optimal coverage node distribution created by GA can be used in a distributed scenario. Ours works are simulated on JGAP and NS2. The simulation result shows that our partition algorithm based on coverage compensation can achieve the same performance as OCOPS in terms of coverage and number of active sensors while using less control messages.

Energy efficiency and low message delay are the basic QoS requirements for data collection in vehicular ad-hoc and sensor networks. In order to achieve these requirements, many solutions have been developed by attaching collectors/sinks to vehicle or aircraft and moving them either randomly or along a predefined route. This paper presents an adaptive data gathering protocol (ADG) that employs multiple mobile collectors (instead of sinks) to help an existing wireless sensor network achieve such requirements. In the proposed ADG protocol, a virtual elastic-force model is used to help mobile collectors adjust their moving speed and direction while adapting to changes within the network. Because of the irregularity of the information generation rate as well as the cost of mobile collectors, the number of collectors can not be predefined. Mobile collectors are sent out by sink if the information in the network is beyond the capabilities of existing mobile collectors, and are called back when they become redundant. In this paper we discuss our ADG protocol, highlight how its performance can be enhanced using both multi-hop OCOPS and LEACH protocols. We report on the performance evaluation of ADG protocol using an extensive set of simulation experiments. Our simulation results show that the proposed ADG algorithm can be a flexible means of meeting the constraints and different requirements of monitoring applications.

The Voronoi Diagram(VD) is a fundamental algorithm for resolving the coverage problems of wireless sensor networks. Most VD based coverage algorithms collect global location information and construct a VD using a centralized construction algorithm. However, the collection process of global location information is expensive and can not be guaranteed in all scenarios. Based on a localized VD construction algorithm this paper proposes a region coverage algorithm and a fully sponsored coverage algorithm where the GPS requirement for global location awareness is removed. The performance of such algorithms is evaluated by the complexity and message cost through a set of simulation experiments. The results show that our algorithms are better in both aspects when they are compared to the centralized region coverage algorithms and central angle algorithm. The performance and flexibility of our algorithms make them a potential solution for regional coverage and off- duty applications.

To prolong the lifetime of wireless sensor networks, a duty cycle scheme is used in energy efficient coverage protocols. By turning off redundant sensors and waking them up when network coverage is jeopardized, the network lifetime is extended while network coverage is preserved. However, it was observed that when the network density is increased to further prolong lifetime, the lifetime increases more slowly than the increase in density. Based upon an analysis of energy usage and energy distribution of the original duty-cycle scheme, we present a new coverage protocol based on function layers that divides nodes into sensing and transmission virtual layers to change the distribution of energy usage in the different layers. The performance of our protocol was compared to the original duty-cycle scheme in terms of energy saving and network lifetime with a high coverage rate.

To prolong the lifetime of wireless sensor network an energy-efficient protocol is required, what can overcome the shortage of energy and minimized the energy consumption while satisfying given throughput and delay requirements. We described a virtual MIMO system by borrowing the ideas of MIMO systems which promises performance enhancements over conventional SISO systems for the same transmission power consumption. Based upon an energy-efficiency analysis of virtual MIMO system we present a self-organize MIMO protocol which described a process of establishing a multi-hop virtual MIMO system based on an existed wireless sensor network. The performance of virtual MIMO system was compared with original wireless sensor network in terms of energy consumption of transmitting a message between two sensors through a multi-hop path.

Coverage is an important issue related to WSN quality of service. Several centralized/decentralized solutions based on the geometry information of sensors and under the assumption of the disk sensing range have been introduced in the literature. However, the disk sensing range assumption is too strong for applications in the real world and cannot be held in required high accurate scenarios, such as the emergency preparedness class of applications. This paper proposes a new intersection point method (IPM) that extends the disk sensing range assumption to an irregular simple polygon assumption. A unit circle test method was also devised in order to provide a controllable degree of accuracy in the determination of fully covered nodes. By adjusting the radius "r" of this unit circle test the algorithm can be made tolerant to holes of a certain size. This makes the solution flexible when the degree of accuracy must be controlled. IPM performance was evaluated through a set of simulation experiments implemented in the NS-2 simulator. Those results were compared to the results for the Central Angle Method (CAM)-part of the C-PNSS scheme[6], and the Association Sponsors Method (ASM)-part of the OCoPS solution[1]. The results show that under the simple polygon sensing range assumption our solution can efficiently identify fully covered sensors, discover holes (blind points), and archive better quality results than CAM and ASM. The performance and flexibility of IPM makes it a potential solution for applications that require a high coverage rate with controllable hole tolerance.

Coverage preserving solutions have been reported and are basically based upon node scheduling and quality coverage. Node scheduling based solutions presents at least three challenging problems: 1) resolving conflicts when determining what nodes should be turned-off to save energy; 2) finding optimal wakeup strategies that avoid waking up more nodes than necessary; and 3) keeping connectivity and coverage of the network while optimizing the number of nodes. This paper presents a novel distributed solution that extends the coverage calculation method proposed by Tian and Georganas, referred here as C-PNSS scheme. The solution is based on local information exchange without the uncertainty of self-schedule algorithms. Our Optimal Coverage-Preserving Scheme (OCoPS) comprises a new Decision algorithm, devised to resolve the off-duty conflict problem for different network densities, and an energy aware Wakeup scheme that solves problems, such as coverage holes, that exist in current schemes. We discuss the implementation of OCoPS with the new Decision algorithm as an extension of LEACH. A set of simulation experiments was performed to evaluate OCoPS performance when compared to the C-PNSS scheme. Our results indicate that our solution outperforms C-PNSS by over 20% on network lifetime and by over 25% on network lifetime when the coverage rate is higher than 80%. The experimental results also show that our coverage scheme based on our extended coverage calculation method effectively limits the on-duty node number when compared the original C-PNSS scheme.

In emergency preparedness, the supervision and control of context aware physical environments that can be subjected to critical conditions, such as fire or leaking of toxic liquids, have to be well covered and reliably monitored by a wireless sensor network. In this paper, we present an efficient energy and coverage protocol using wireless sensors and discuss its energy cost analysis