PSNR 29.47 PSNR 28.80 PSNR 27.28 PSNR 30.84 PSNR 26.69 PSNR 26.

High-quality large-scale scene rendering requires a scalable representation and accurate camera poses. This research combines tile-based hybrid neural fields with parallel distributive optimization to improve bundle-adjusting neural radiance fields. The proposed method scales with a divide-and-conquer strategy. We partition scenes into tiles, each with a multi-resolution hash feature grid and shallow chained diffuse and specular multilayer perceptrons (MLPs). Tiles unify foreground and background via a spatial contraction function that allows both distant objects in outdoor scenes and planar reflections as virtual images outside the tile. Decomposing appearance with the specular MLP allows a specular-aware warping loss to provide a second optimization path for camera poses. We apply the alternating direction method of multipliers (ADMM) to achieve consensus among camera poses while maintaining parallel tile optimization. Experimental results show that our method outperforms state-of-the-art neural scene rendering method quality by 5%--10% in PSNR, maintaining sharp distant objects and view-dependent reflections across six indoor and outdoor scenes.

We propose a scalable neural scene reconstruction and rendering method to support distributed training and interactive rendering of large indoor scenes. Our representation is based on tiles. Tile appearances are trained in parallel through a background sampling strategy that augments each tile with distant scene information via a proxy global mesh. Each tile has two low-capacity MLPs: one for view-independent appearance (diffuse color and shading) and one for view-dependent appearance (specular highlights, reflections). We leverage the phenomena that complex view-dependent scene reflections can be attributed to virtual lights underneath surfaces at the total ray distance to the source. This lets us handle sparse samplings of the input scene where reflection highlights do not always appear consistently in input images. We show interactive free-viewpoint rendering results from five scenes, one of which covers an area of more than 100 m 2 . Experimental results show that our method produces higher-quality renderings than a single large-capacity MLP and five recent neural proxy-geometry and voxel-based baseline methods. Our code and data are available at project webpage https://xchaowu.github.io/papers/scalable-nisr.

This paper proposes a novel scalable image-based rendering (IBR) pipeline for indoor scenes with reflections. We make substantial progress towards three sub-problems in IBR, namely, depth and reflection reconstruction, view selection for temporally coherent view-warping, and smooth rendering refinements. First, we introduce a global-mesh-guided alternating optimization algorithm that robustly extracts a two-layer geometric representation. The front and back layers encode the RGB-D reconstruction and the reflection reconstruction, respectively. This representation minimizes the image composition error under novel views, enabling accurate renderings of reflections. Second, we introduce a novel approach to select adjacent views and compute blending weights for smooth and temporal coherent renderings. The third contribution is a supersampling network with a motion vector rectification module that refines the rendering results to improve the final output's temporal coherence. These three contributions together lead to a novel system that produces highly realistic rendering results with various reflections. The rendering quality outperforms state-of-the-art IBR or neural rendering algorithms considerably.

Vector image is a type of image composed of many geometric primitives. Compared with bitmaps, vector images have the ability to save memory as well as to enlarge without distortion. Meanwhile, it has been commonly adopted in data visualization (image data) because it can be scaled to multiple sizes to fit different scenes. For instance, it can be applied for the illustrations in newspapers and magazines, the logo on the web, the background for poster, the design of text, and traffic signs. However, transforming a bitmap to vector image is still a challenging problem because of the complicated content of a bitmap, which tends to consist of more than just simple geometry. Aiming at this issue, there is a new approach proposed to transform from bitmaps to vector images, which is based on triangle units and consists of three steps. In detail, firstly, there is an initial mesh constructed for one image in pixel level after detecting features. Then, the initial mesh will be simplified by collapsing two vertices as the initial mesh is too dense to represent one image. Specifically, there are two main parts in this step, which are collapse conditions and collapse influences. In the step of collapsing, issues such as overlap and sharp triangles can be conquered by a sort-edge method (which will be illustrated specifically later). The final step is to select one color for each triangle, since it is helpful to save the memory and speed up the process of this method. In addition, one color will represent one triangle; hence, in the final step, the four-triangle sample method will be applied in order to prevent a vector image from generating too large color discontinuity. Once the pretest proceeds without mistake, the method above is able to work for the general bitmaps. Note that our method can be applied to information security and privacy, since one image can be encoded to some triangles and colors.

In many emerging wireless sensor network scenarios, the use of a fixed infrastructure of base stations for data collection is either infeasible, or prohibitive in terms of deployment and maintenance costs. Instead, we consider the use of mobile devices (i.e. smartphones) carried by people in their daily life to collect data from sensor nodes opportunistically. As the movement of these mobile nodes is, by definition, not controlled for the purpose of data collection, synchronization through contact probing becomes a challenging task, particularly for sensor nodes, which need to be aggressively duty-cycled to conserve energy and achieve long lifetimes. This paper formulates this important problem, providing an analytical solution framework and systematically investigating the effective use of contact probing for opportunistic data collection. We present two new solutions, Sensor Node-Initiated Probing (SNIP) and SNIP-Rush Hours, the latter taking advantage of the temporal locality of human mobility. These schemes are evaluated using numerical analysis and COOJA network simulations, and the results are validated on a small sensor testbed and with the real-world human mobility traces from Nokia MDC Dataset. Our experimental results quantify the relative performance of alternative solutions on sensor node energy consumption and the efficacy of contact probing for data collection, allowing us to offer insights on this important emerging problem.

XG-PON is the next-generation standard for passive optical networks operating at 10Gb/s and TCP is the dominant transport protocol of the Internet. In this paper, we present the first performance evaluation of TCP over XG-PON, considering efficiency, fairness, responsiveness, and convergence. The impact of XG-PON's large delay-bandwidth product and asymmetric bandwidth provision are assessed, together with the dynamic bandwidth allocation mechanism. Our state-of-the-art NS3 simulation uses real implementations of three TCP variants (Reno, CUBIC and H-TCP) from the Network Simulation Cradle. Our results highlight several issues that arise for TCP over XG-PON, and emphasise the need for improved awareness of medium access control and scheduling in the context of specific TCP congestion control behaviour.

Opportunistic data collection in wireless sensor networks uses passing smartphones to collect data from sensor nodes, thus avoiding the cost of multiple static sink nodes. Based on the observed mobility patterns of smartphone users, sensor data should be preforwarded to the nodes that are visited more frequently with the aim of improving network throughput. In this article, we construct a formal network model and an associated theoretical optimization problem to maximize the throughput subject to energy constraints of sensor nodes. Since a centralized controller is not available in opportunistic data collection, data pre-forwarding (DPF) must operate as a distributed mechanism in which each node decides when and where to forward data based on local information. Hence, we develop a simple distributed DPF mechanism with two heuristic algorithms, implement this proposal in Contiki-OS, and evaluate it thoroughly. We demonstrate empirically, in simulations, that our approach is close to the optimal solution obtained by a centralized algorithm. We also demonstrate that this approach performs well in scenarios based on real mobility traces of smartphone users. Finally, we evaluate our proposal on a small laboratory testbed, demonstrating that the distributed DPF mechanism with heuristic algorithms performs as predicted by simulations, and thus that it is a viable technique for opportunistic data collection through smartphones.

As bandwidth in the Internet continues to grow, there will be more and more long fat network pipes with abundant residual bandwidth. At the same time, there is also a gradual and steady increase in the deployment of Internet endpoints equipped with different variants of high speed TCP. In this work, we first illustrate drawbacks associated with two widely deployed high speed TCP variants, namely: cubic TCP and compound TCP. We show that even with common and reasonable settings, problems can arise. Next, we present synchronized TCP (Sync-TCP), a new delay-based high speed congestion control (HSCC) algorithm. The approach taken by Sync-TCP is novel in the following ways. First, Sync-TCP exploits synchronization. The key insight of Sync-TCP is that if competing flows could detect the same congestion signal through queue delay, these flows can coordinate their congestion control behaviors. Second, using only the basic mechanism, Sync-TCP will yield to legacy TCP when congestion is detected. Hence, Sync-TCP is designed to not hurt applications using legacy TCP or interactive applications. We performed extensive simulation and some testbed evaluations to show that Sync-TCP achieves its design goals and it performs better than existing HSCC approaches including Fast TCP, Compound TCP and Cubic TCP, especially in the trade-off between throughput and friendliness.

In recent years, many high-speed congestion control (HSCC) algorithms have been proposed for utilizing network pipes with huge bandwidth-delay product (BDP), and some of them have also been implemented in popular operating systems. Considering the extremely long round trip propagation delay (RTPD) of satellite network, it is very likely that these algorithms are triggered when TCP flows pass through satellite network. But the existing algorithms are normally evaluated on network pipes with high bandwidth and moderate RTPD (< 400 ms). This paper is an attempt to study these algorithms on a simulated satellite network with moderate bandwidth and extremely long RTPD. Their effects on the existing applications, especially World Wide Web (WWW) and the emerging streaming applications, are emphasized. Different queue sizes are also used in simulation with the aim of investigating how to provision satellite link's queue for well accommodating flows driven by HSCC algorithms. Through this study, we find that Compound-TCP, Cubic-TCP, and H-TCP can not work well on satellite network. Currently, satellite link should use moderate queue size for accommodating flows driven by HSCC algorithms. As for end hosts, when they find that round trip time (RTT) is very long, HS-TCP should be adopted for high throughput while avoiding to hurt the existing applications as little as possible.

In recent years, many different kinds of wireless access networks have been deployed for the Internet and have become inseparable parts of the Internet. But TCP, the most widely used transport protocol of the Internet, was designed for stationery hosts. In particular, TCP faces severe challenges when user moves around in these networks and handoff occurs frequently. In this paper, TCP handoff (TCP-HO), a practical end-to-end mechanism, is proposed for improving TCP performance in heterogeneous mobile environments. TCP-HO assumes that a mobile host is able to detect the completion of handoff immediately and has a coarse estimation of new wireless link's bandwidth. When a mobile host detects handoff completion, it will immediately notify the server through two duplicate ACKs, whose TCP option also carries the bandwidth of new wireless link. After receiving this notification, the server begins to transmit immediately and keeps updating ssthresh according to the bandwidth from mobile host and its new RTT samples. This updating will be stopped after four RTT samples or after congestion is detected. TCP-HO has been implemented in FreeBSD 5.4. Experimental results show that TCP-HO does improve TCP performance without adversely affecting cross traffic in a heterogeneous mobile environment.

A wireless sensor network is normally a multi-hop wireless ad hoc network whose purpose is to transmit events sensed by sensors to the sink. Since sensor nodes are normally powered by non-rechargeable and non-replaceable battery, an energy efficient event transport service is crucial for extending the life time of a wireless sensor network. Due to the possibility of crossing multiple lossy links between the source and the sink, many events may be lost before they arrive at the sink. In addition, the overhead of low layer protocols is very large due to the short event data. These two factors can result in significant event loss and waste of energy. Many mechanisms have been proposed on other networks to improve reliability and reduce overhead. Due to scarce memory and poor computing ability of sensor nodes in a sensor network, these mechanisms may not be feasible to implement and they may not work efficiently with respect to energy. In this paper, PMC (Pipelined Multiple Copies) is proposed for transmitting events in a wireless sensor network, with the goals of achieving low event loss rate and high energy efficiency. In PMC, multiple copies of an event are pipelined and transmitted in several consecutive frames according to Silent CSMA, a variant of CSMA/CA. Multiple copies of an event reduce end-to-end event loss rate and Silent CSMA reduces packet energy consumption significantly.

TCP, perhaps the most widely used transport protocol, was designed for wired links and stationary hosts. But more and more links with different characteristics are used to access the Internet. The variations in link characteristics present different problems to TCP and enhancement of TCP performance over these links remains an active area of research. Since the current core network of the Internet is composed mainly of optical fiber links and high speed routers, the characteristics of the access link (or the last link) usually dominate the characteristics of the end-to-end network path. As a result, if the client can acquire the characteristics of the access link (e.g. through measurement), the TCP server can utilize heuristics that exploit the access link's characteristics in order to improve the performance of TCP. In this paper, we propose and evaluate the performance of several heuristics that exploit knowledge of access link characteristics such as bandwidth and packet error rate. Simulation results show that these heuristics can indeed improve TCP performance significantly and demonstrate the potential benefits of using heuristics that adjust TCP congestion control behaviors to match the access link characteristics.

Knowledge of network path characteristics enables upper layers to improve performance through adaptation protocols which adjust behaviors according to current network path characteristics. For a mobile node of a wireless network, this knowledge is especially valuable because the characteristics of the wireless link, which often dominate network path characteristics, change frequently and abruptly due to mobility, fading, and contention among nodes. Due to the dynamics and precious bandwidth of the wireless link, the commonly used probing packet methods are not appropriate for estimating the characteristics of a network path with a wireless link. We propose a new mechanism to estimate link characteristics of IEEE 802.11 DCF (distributed coordination function) based wireless LAN (WLAN), one of the most popular wireless access networks. Instead of sending probing-packets, a mobile node estimates its link characteristics based on wireless link quality and the contention status of the whole WLAN. We implement this mechanism and test its accuracy in NS2.