Cross entropy is the most widely used loss function for supervised training of image classification models. In this paper, we propose a novel training methodology that consistently outperforms cross entropy on supervised learning tasks across different architectures and data augmentations. We modify the batch contrastive loss, which has recently been shown to be very effective at learning powerful representations in the self-supervised setting. We are thus able to leverage label information more effectively than cross entropy. Clusters of points belonging to the same class are pulled together in embedding space, while simultaneously pushing apart clusters of samples from different classes. In addition to this, we leverage key ingredients such as large batch sizes and normalized embeddings, which have been shown to benefit self-supervised learning. On both ResNet-50 and ResNet-200, we outperform cross entropy by over 1%, setting a new state of the art number of 78.8% among methods that use AutoAugment data augmentation. The loss also shows clear benefits for robustness to natural corruptions on standard benchmarks on both calibration and accuracy. Compared to cross entropy, our supervised contrastive loss is more stable to hyperparameter settings such as optimizers or data augmentations.

Panorama creation is one of the most widely deployed techniques in computer vision. In addition to industry applications such as Google Street View, it is also used by millions of consumers in smartphones and other cameras. Traditionally, the problem is decomposed into three phases: registration, which picks a single transformation of each source image to align it to the other inputs, seam finding, which selects a source image for each pixel in the final result, and blending, which fixes minor visual artifacts [1, 2]. Here, we observe that the use of a single registration often leads to errors, especially in scenes with significant depth variation or object motion. We propose instead the use of multiple registrations, permitting regions of the image at different depths to be captured with greater accuracy. MRF inference techniques naturally extend to seam finding over multiple registrations, and we show here that their energy functions can be readily modified with new terms that discourage duplication and tearing, common problems that are exacerbated by the use of multiple registrations. Our techniques are closely related to layer-based stereo [3, 4], and move image stitching closer to explicit scene modeling. Experimental evidence demonstrates that our techniques often generate significantly better panoramas when there is substantial motion or parallax.

Google started the WebM Project in 2010 to develop open source, royalty- free video codecs designed specifically for media on the Web. The second generation codec released by the WebM project, VP9, is currently served by YouTube, and enjoys billions of views per day. Realizing the need for even greater compression efficiency to cope with the growing demand for video on the web, the WebM team embarked on an ambitious project to develop a next edition codec AV1, in a consortium of major tech companies called the Alliance for Open Media, that achieves at least a generational improvement in coding efficiency over VP9. In this paper, we focus primarily on new tools in AV1 that improve the prediction of pixel blocks before transforms, quantization and entropy coding are invoked. Specifically, we describe tools and coding modes that improve intra, inter and combined inter-intra prediction. Results are presented on standard test sets.

Event-driven wireless sensor networks (WSNs) usually operate under light traffic load. However, when an event is detected, a large number of packets may be generated. A MAC protocol designed for this kind of WSNs should be able to swiftly adapt to the two conditions. Most WSN MAC protocols are optimized for light traffic for the energy efficiency consideration. In this paper, we propose a novel receiver-centric MAC protocol called RC-MAC that seamlessly integrates duty cycling and receiver-centric scheduling, providing high throughput without sacrificing the energy efficiency. To handle bursty traffic triggered by an event, RC-MAC takes advantage of the underlying data gathering tree structure of WSNs and the multichannel technique supported by current IEEE 802.15.4 RF transceivers to assist scheduling of medium access. The throughput is improved in two phases with receiver-centric medium access scheduling and distributed channel assignment. First, on a data gathering tree, a receiver is able to coordinate the medium access of multiple senders so as to reduce collisions and achieve high throughput. Second, different receivers coordinate their senders in different channels and the throughput is further improved by allowing parallel data gathering. Observing packet processing time on low cost sensor nodes, we design a scheduling pattern that ensures fairness between source nodes without sacrificing the throughput. We evaluate the performance of our RC-MAC through measurements of an implementation in TinyOS on TelosB motes and extensive ns-2 simulations. Compared with contention-based and scheduling-based MAC protocols, we show that the throughput and the fairness under heavy traffic load are significantly improved by the receiver-centric scheduling. Due to the high throughput, the energy efficiency is also improved.

Piecewise planar model for stereo matching can overcome the challenges presented by poorly textured surfaces. Lots of works employ color segmentation cues to build piecewise planar model. However, segmentation is not sufficient to represent the content-consistency, since segments are usually too small or too messy to ensure pixels from the same object surface to be assigned to the same disparity layer. To obtain more compact piecewise planar models for urban scenes, in this work, a two-layer (pixel-wise and semantic-piecewise) graph structure, which incorporates cues from image segmentation and semantic classification, is proposed. One of the graph layers models pixel color-consistency in image pairs, and the other models the disparity smoothness between image segments of the same object according to semantic classification. Experiments on different urban scenes justify the efficiency of our method.

In this paper, a beamforming (BF) based cooperation scheme with decode-and-forward (DF) protocol is proposed and its performance analysis is provided. Firstly, to solve the issue of carrier synchronization in distributed networks, a backward synchronization (B-Sync) method is introduced, where participating relays synchronize to the reference carrier broadcast by the destination and make signals superimposed coherently at the destination without any phase feedback. Then, based on B-sync, a cooperation scheme with DF protocol is proposed to carry out BF forwarding among relays. The outage probability of this scheme is derived and analyzed, which shows that the full diversity order can be achieved by the proposed scheme. Furthermore, symbol error probability (SEP) analysis is presented to evaluate the practical performance. Simulation results validate the analysis, and show that the proposed scheme outperforms other existing ones in terms of outage probability and SEP.

Greedy forwarding is a simple yet efficient technique employed by many routing protocols. It is ideal to realize point-to-point routing in wireless sensor networks because packets can be delivered by only maintaining a small set of neighbors' information regardless of network size. It has been successfully employed by geographic routing, which assumes that a packet can be moved closer to the destination in the network topology if it is forwarded geographically closer to the destination in the physical space. This assumption, however, may lead packets to the local minimum where no neighbors of the sender are closer to the destination or low-quality routes that comprise long distance hops of low packet reception ratio. To address the local minimum problem, we propose a topology aware routing (TAR) protocol that efficiently encodes a network topology into a low-dimensional virtual coordinate space where hop distances between pairwise nodes are preserved. Based on precise hop distance comparison, TAR can assist greedy forwarding to find the right neighbor that is one hop closer to the destination and achieve high success ratio of packet delivery without location information. Further, we improve the routing quality by embedding a network topology based on the metric of expected transmission count (ETX). ETX embedding accurately encodes both a network's topological structure and channel quality to nodes' small size virtual coordinates, which helps greedy forwarding to guide a packet along the optimal path that has the fewest number of transmissions. We evaluate our approaches through both simulations and experiments, showing that routing performance are improved in terms of routing success ratio and routing cost.

Using stereo disparity or depth information to detect and track moving objects is receiving increasing attention in recent years. However, this approach suffers from some difficulties, such as synchronisation between two cameras and doubling of the image-data size. Besides, traditional stereo-imaging systems have a limited field of view (FOV), which means that they need to rotate the cameras when an object moves out of view. In this research, the authors present a depth-space partitioning algorithm for performing object tracking using single-camera omni-stereo imaging system. The proposed method uses a catadioptric omni-directional stereo-imaging system to capture omni-stereo image ‘pairs.’ This imaging system has 360° FOV, avoiding the need for rotating cameras when tracking a moving object. In order to estimate omni-stereo disparity, the authors present a depth-space partitioning strategy. It partitions three-dimensional depth space with a series of co-axial cylinders, models the disparity estimation as a pixel-labelling problem and establishes an energy minimisation function for solving this problem using graph cuts optimisation. Based on the omni-stereo disparity-estimation results, the authors detect and track-moving objects based on omni-stereo disparity motion vector, which is the difference between two consecutive disparity maps. Experiments on moving car tracking justify the proposed method.

We consider a two-way relay network consisting of multiple pairs of single-antenna users and multiple distributed single-antenna relays. The two communication peers in each pair of users transmit simultaneously to the relays in the first time slot, and the relays rebroadcast the received sum signal weighted by a complex gain, in the second time slot. For multi-user systems, the signal arriving at the users contains not only self interference from the back-propagation of user signals, but also inter-pair interferences from other pairs of users. In this paper, we use zero-forcing (ZF) to cancel the inter-user interference, assuming that channel-state information for all relay-peer connections are known at every relay, but no data exchange occurs between relays. We also derive two closed-form expressions for zero-forcing beamforming weights, corresponding to two different relay power constraints, which can be implemented in a distributed manner. The first approach uses standard ZF to null out every inter-pair interference and the second approach sets the total inter-pair interference to zero. We also derive a closed-form upper bound of the achievable sum-rate and show that both methods achieve the same multiplexing gain when the number of relays N is sufficient for perfect zero-forcing, namely 2K2 + K, where K is the number of user pairs. For the case of insufficient number of relays, we also propose two solutions for beamforming weights, i.e., based on diagonal loading and use of the pseudo-inverse, and compare their advantages and weaknesses.

Node localization plays an important role in many practical applications of wireless underground sensor networks (WUSNs), such as finding the locations of earthquake epicenters, underground explosions, and microseismic events in mines. It is more difficult to obtain the time-difference-of-arrival (TDOA) measurements in WUSNs than in terrestrial wireless sensor networks because of the unfavorable channel characteristics in the underground environment. The robust Chinese remainder theorem (RCRT) has been shown to be an effective tool for solving the phase ambiguity problem and frequency estimation problem in wireless sensor networks. In this paper, the RCRT is used to robustly estimate TDOA or range difference in WUSNs and therefore improves the ranging accuracy in such networks. After obtaining the range difference, distributed source localization algorithms based on a diffusion strategy are proposed to decrease the communication cost while satisfying the localization accuracy requirement. Simulation results confirm the validity and efficiency of the proposed methods.

An omni-directional stereo system has a wider field-of-view than conventional cameras, and is widely used in many applications such as robots navigation, depth estimation, and 3D reconstruction. Existing approaches usually use single viewpoint (SVP) systems as the imaging sensor. However, literature proves that an efficient SVP of an omni-directional system can only be achieved with precisely aligned mirrors of parabolic or hyperbolic profile. This enforces rigorous restriction on the configuration of camera and mirrors. In fact, some other profiles, though they do not have the SVP property, are desirable for certain reasons such as cheaper cost and more practical implementation. Therefore, in this paper, we propose both a typical nonsingle viewpoint (non-SVP) omni-directional stereo sensor and its corresponding depth estimation method based on graph-cuts optimization. The sensor comprises a perspective camera and two separate reflective mirrors that could be any radially-symmetric ones. To formulate the depth estimation more consistent with the proposed sensor, we divide the depth space of scenes with a sequence of virtual coaxial cylindrical layers, and model depth estimation as a labeling problem. In the labeling procedure, by considering the characteristics of an omni-directional image, we further devise novel tangential-neighborhood system, radial-neighborhood system, and depth-gradual-changing smoothness constraint which perform better than traditional ones. Depth estimation and 3D reconstruction for both synthesis and real scenes justify the effectiveness of the proposed method.

Distributed beamforming is a form of collaborative communication, in which wireless nodes adjust the phases of their transmitted signals so that the signals are combined coherently at the destination. This enhances the received signal power and extends the communication range. In this paper, we propose a transmit beamforming method that does not require any explicit feedback information from the receiver for the phase adjustment. Instead, sensors use phase-locked loops (PLL) to synchronize their carriers to the received reference signal broadcast by the destination / receiver. Based on the fact that the change of phase over some distance depends not only on the distance but also on the frequency, the destination indirectly adjusts the sensors' phase by changing the frequency of the reference signal. The destination scans all the frequencies within a feasible range to find the optimal frequency at which the received power gets its maximum. The advantage of this method is that it does not require any direct interaction among the sensors and feedback between the sensors and the destination. Every sensor adjusts its own phase independently and simultaneously. Therefore, the proposed method has good scalability and is suitable for the distributed system. We also derive some performance analysis of the method and present simulation results that validate the analysis.

The Chinese remainder theorem (CRT) is an effective tool to solve the phase ambiguity problem in phase-based range estimation. However, existing methods suffer from problems such as requiring special measuring frequency, low spectrum efficiency, noise sensitivity, etc. To overcome these problems, this paper presents a CRT ranging method using two "adjacent" frequencies. As a result, all the frequencies in the given frequency span can be used, which greatly increases spectrum utilization. Moreover, since the same distance is measured by different frequency pairs, statistical processing can be performed on the results, which further improves estimation accuracy. Simulations verify the validity of the proposed method.

RIPS系统通过测量干涉信号的相位提供了一种精度高、设备简单的无线传感器网络定位方法.但利用相位进行测距和定位的方法不可避免地存在相位模糊问题,RIPS系统采取的方式是在多个频率下对某一距离差进行相位测量,根据各相位测量值来搜索实际距离差.为了避免RIPS系统中这种繁复的搜索过程,文中提出了一种高效的距离差估计方法.该方法基于中国余数定理（CRT）通过闭式一次直接计算得出距离估计值,从而避免了搜索过程,极大地降低了节点运算能耗并且提高了定位系统的实时性.同时,为了克服传统CRT算法对噪声敏感的缺点,利用算法中加权系数具有的一些性质来减小噪声对估计结果的影响,提高了算法的鲁棒性.仿真结果表明该方法鲁棒性好,估计精度高,在无线传感器网络中具有良好的可行性.

The radio interferometric positioning system (RIPS) measures the phase difference of the interference signal to provide high accuracy and at the same time maintain simple hardware configuration for wireless sensor networks. However, it suffers from phase ambiguity problem because of the periodicity of phase, which makes it hard to determine the actual distance difference only from a single phase measurement. To solve the problem, RIPS makes multiple measurements at different frequencies so as to determine the distance difference from multiple phases. However, this is a computationally intensive searching process and not suitable for energy-constrained wireless sensor nodes. In this paper, we introduce the Chinese remainder theorem (CRT) to RIPS to solve the phase ambiguity problem. Meanwhile, we utilize some properties of the coefficients in the CRT algorithm to avoid the over-sensitivity of the traditional CRT, which increases the robustness of the algorithm. We apply this robust CRT algorithm to the ranging process which calculates the distance difference directly from a closed-form equation and therefore reduces the response time and the energy consumption of the ranging procedure.

The Radio Interferometric Positioning System (RIPS) provides a localization method with high accuracy and simple hardware configuration for sensor networks by means of measuring the phase difference of the interference signal. However, due to the periodicity of phase, it is hard to determine the actual distance difference only from a single phase measurement. To solve the problem, RIPS makes multiple measurements at different frequencies so as to convert phase difference to distance difference, which is a computationally intensive searching process and not suitable for the energy-constrained wireless sensor nodes. In this paper, we propose an efficient ranging method based on Chinese Remainder Theorem (CRT). Meanwhile, we utilize some properties of the coefficients in the CRT algorithm to avoid the over-sensitivity of the traditional CRT, which increases the robustness of the algorithm. We apply this robust CRT algorithm to the ranging process which calculates the distance difference directly in a closed form equation and therefore reduces the response time and energy consumption of the ranging procedure.

Phase feedback and adjustment between wireless nodes greatly reduce the power efficiency of distributed beamforming. In this paper, we propose a distributed transmit beamforming method without any phase feedback between nodes. The concept of our approach is to have the received signals retrace their ways, so that the phase offset of the forward path compensates that of the backward path; as a result, signals from different nodes in-phase combine at the destination. Therefore, the received power or the communication range is increased. In order to implement the concept of "retracement", we also propose a transceiver prototype which is based on the Direct Digital Synthesis technique. Experimental and simulation results validate the effectiveness of our approach.

Locating sensors in an indoor environment is a challenging problem due to the insufficient distance measurements caused by short ultrasound range and the incorrect distance measurements caused by multipath effect of ultrasound. In this paper, we propose a virtual ruler approach, in which a vehicle equipped with multiple ultrasound beacons travels around the area to measure distances between pairwise sensors. Virtual Ruler can not only obtain sufficient distances between pairwise sensors, but can also eliminate incorrect distances in the distance measurement phase of sensor localization. We propose to measure the distance between pairwise sensors from multiple perspectives using the virtual ruler and filter incorrect values through a statistical approach. By assigning measured distances with confidence values, the localization algorithm can intelligently localize each sensor based on high confidence distances, which greatly improves localization accuracy. Our performance evaluation shows that the proposed approach can achieve better localization results than previous approaches in an indoor environment.