Academic literature on the topic 'Full-initialization Points'

Abstract. Although a number of emerging point-cloud semantic segmentation methods achieve state-of-the-art results, acquiring fully interpreted training data is a time-consuming and labor-intensive task. To reduce the burden of data annotation in training, semiand weakly supervised methods are proposed to address the situation of limited supervisory sources, achieving competitive results compared to full supervision schemes. However, given a fixed budget, the effective annotation of a few points is typically ignored, which is referred to as weak-label initialization in this study. In practice, random selection is typically adopted by default. Because weakly supervised methods largely rely on semantic information supplied by initial weak labels, this studies explores the influence of different weak-label initialization strategies. In addition to random initialization, we propose a feature-constrained framework to guide the selection of initial weak labels. A feature space of point clouds is first constructed by feature extraction and embedding. Then, we develop a density-biased strategy to annotate points by locating highly dense clustered regions, as significant information distinguishing semantic classes is often concentrated in such areas. Our method outperforms random initialization on ISPRS Vaihingen 3D data when only using sparse weak labels, achieving an overall accuracy of 78.06% using 1‰ of labels. However, only a minor increase is observed on the LASDU dataset. Additionally, the results show that initialization with category-wise uniformly distributed weak labels is more effective when incorporated using a weakly supervised method.

This paper demonstrates that multilingual denoising pre-training produces significant performance gains across a wide variety of machine translation (MT) tasks. We present mBART—a sequence-to-sequence denoising auto-encoder pre-trained on large-scale monolingual corpora in many languages using the BART objective (Lewis et al., 2019 ). mBART is the first method for pre-training a complete sequence-to-sequence model by denoising full texts in multiple languages, whereas previous approaches have focused only on the encoder, decoder, or reconstructing parts of the text. Pre-training a complete model allows it to be directly fine-tuned for supervised (both sentence-level and document-level) and unsupervised machine translation, with no task- specific modifications. We demonstrate that adding mBART initialization produces performance gains in all but the highest-resource settings, including up to 12 BLEU points for low resource MT and over 5 BLEU points for many document-level and unsupervised models. We also show that it enables transfer to language pairs with no bi-text or that were not in the pre-training corpus, and present extensive analysis of which factors contribute the most to effective pre-training. 1

Simultaneous localization and mapping (SLAM) are fundamental elements for many emerging technologies, such as autonomous driving and augmented reality. For this paper, to get more information, we developed an improved monocular visual SLAM system by using omnidirectional cameras. Our method extends the ORB-SLAM framework with the enhanced unified camera model as a projection function, which can be applied to catadioptric systems and wide-angle fisheye cameras with 195 degrees field-of-view. The proposed system can use the full area of the images even with strong distortion. For omnidirectional cameras, a map initialization method is proposed. We analytically derive the Jacobian matrices of the reprojection errors with respect to the camera pose and 3D position of points. The proposed SLAM has been extensively tested in real-world datasets. The results show positioning error is less than 0.1% in a small indoor environment and is less than 1.5% in a large environment. The results demonstrate that our method is real-time, and increases its accuracy and robustness over the normal systems based on the pinhole model. We open source in https://github.com/lsyads/fisheye-ORB-SLAM.

The radar-enhanced GSI (version 3.1) system and the WRF-ARW (version 3.4.1) model were modified to assimilate radar/lightning-proxy reflectivity. First, cloud-to-ground lightning data were converted to reflectivity using a simple assumed relationship between flash density and reflectivity. Next, the reflectivity was used in the cloud analysis of GSI to adjust the cloud/hydrometeors and moisture. Additionally, the radar/lightning-proxy reflectivity was simultaneously converted to a 3D temperature tendency. Finally, the model-calculated temperature tendencies from the explicit microphysics scheme, as well as cumulus parameterization at 3D grid points at which the radar temperature tendency is available, were updated in a forward full-physics step of diabatic digital filter initialization in the WRF-ARW. The WRF-GSI system was tested using a mesoscale convective system that occurred on June 5, 2009, and by assimilating Doppler radar and lightning data, respectively. The forecasted reflectivity with assimilation corresponded more closely to the observed reflectivity than that of the parallel experiment without assimilation, particularly during the first 6 h. After assimilation, the short-range precipitation prediction improved, although the precipitation intensity was stronger than the observed one. In addition, the improvements obtained by assimilating lightning data were worse than those from assimilating radar reflectivity over the first 3 h but improved thereafter.

Abstract. Leads within consolidated sea ice control heat exchange between the ocean and the atmosphere during winter thus constituting an important climate parameter. These narrow elongated features occur when sea ice is fracturing under the action of wind and currents, reducing the local mechanical strength of the ice cover, which in turn impact the sea ice drift patterns. This makes a high quality lead fraction (LF) dataset to be in demand for sea ice model evaluation, initialization and for assimilation of such data in regional models. In this context, the available LF dataset retrieved from satellite passive microwave observations (Advanced Microwave Scanning Radiometer – Earth Observing System, AMSR-E) is of great value, providing pan-Arctic light- and cloud-independent daily coverage since 2002. Here we quantify errors in this dataset using accurate LF estimates retrieved from Synthetic Aperture Radar (SAR) images employing a threshold technique, also introduced in this work. We find a consistent overestimation by a factor of 2–4 of the LF estimates in the AMSR-E LF product. We show for a data sample from the AMSR-E LF dataset that a simple adjustment of the tie points used in the method to estimate the LF can reduce the pixel-wise error by a factor of 2 on average. Applying such adjustment to the full dataset may thus significantly increase the quality and value of the original dataset.

Photovoltaic (PV) modules experience some partial shading conditions (PSC) due to some various factors. In that kind of a condition, a few maximum power points (MPPs) possibly appear on the power-voltage (P-V) curve, which increases the tracking difficulties. It is known that maximum power point tracking (MPPT) may not be realized by hill climbing (HC) based conventional MPPT algorithms under PSCs. In this context, this paper presents a novel micro converter based algorithm that was developed by using P-V characteristics of PV modules. Unlike voltage or duty ratio scanning techniques, this paper introduces a new deciding method to determine the correct global MPP (GMPP) region. For this, the proposed method uses some duty ratios that were calculated corresponding to each MPP region. Thus, the initialization of duty ratio is done properly, which results in high tracking speed and accurate tracking of the GMPP. The other advantages of the proposed algorithm are structural simplicity, less computational burden, and ease of implementation with a basic microcontroller. The simulation results show that this algorithm has fast tracking capability and it manages to track GMPP for PSCs correctly, since it includes an artificial scanning procedure. Single ended primary inductance converter (SEPIC) is built in order to validate the proposed global maximum power point tracking (GMPPT) algorithm. The performance of the proposed GMPPT technique is verified by experimental studies. The results show that the proposed GMPPT technique is fast by up to five times than an adaptive full scanning strategy and improved IC algorithm. Furthermore, the proposed algorithm can be commercially used in micro converters, since it is compatible with small number of bypass diodes in a module.

Indoor environment is a common scene in our everyday life, and detecting and tracking multiple targets in this environment is a key component for many applications. However, this task still remains challenging due to limited space, intrinsic target appearance variation, e.g. full or partial occlusion, large pose deformation, and scale change. In the proposed approach, we give a novel framework for detection and tracking in indoor environments, and extend it to robot navigation. One of the key components of our approach is a virtual top view created from an RGB-D camera, which is named ground plane projection (GPP). The key advantage of using GPP is the fact that the intrinsic target appearance variation and extrinsic noise is far less likely to appear in GPP than in a regular side-view image. Moreover, it is a very simple task to determine free space in GPP without any appearance learning even from a moving camera. Hence GPP is very different from the top-view image obtained from a ceiling mounted camera. We perform both object detection and tracking in GPP. Two kinds of GPP images are utilized: gray GPP, which represents the maximal height of 3D points projecting to each pixel, and binary GPP, which is obtained by thresholding the gray GPP. For detection, a simple connected component labeling is used to detect footprints of targets in binary GPP. For tracking, a novel Pixel Level Association (PLA) strategy is proposed to link the same target in consecutive frames in gray GPP. It utilizes optical flow in gray GPP, which to our best knowledge has never been done before. Then we "back project" the detected and tracked objects in GPP to original, side-view (RGB) images. Hence we are able to detect and track objects in the side-view (RGB) images. Our system is able to robustly detect and track multiple moving targets in real time. The detection process does not rely on any target model, which means we do not need any training process. Moreover, tracking does not require any manual initialization, since all entering objects are robustly detected. We also extend the novel framework to robot navigation by tracking. As our experimental results demonstrate, our approach can achieve near prefect detection and tracking results. The performance gain in comparison to state-of-the-art trackers is most significant in the presence of occlusion and background clutter.

Objectives: Patient-reported outcomes (PROs) are an important clinical tool to effectively collect, compile, and assess patient data and clinical progress. Beyond just tracking patient progress, PROs facilitate patient-provider communication and can be used to inform care practices. Though recent studies have focused on the merits of PRO collection in orthopaedics, few have sought to understand the impact of the coronavirus pandemic on PRO collection. Our objective is to review the initialization of an electronic PRO collection platform at our Orthopaedic Sports Medicine clinics and surgery center before and during the coronavirus pandemic. We focus specifically on our first 7,557 patients in a retrospective study and the interplay between our collection tactics and the patients’ responses to PROs from November 2019 to July 2020. Methods: Starting in the fall of 2019, we implemented a new clinic-wide data capturing system to collect PROs and maintain our outcomes database. We began with a small group of providers and their patients before expanding to include most providers and staff in three clinics. Patients received the following forms prior to their first visit: Patient-Reported Outcome Measurement Information System (PROMIS) Pain Interference, PROMIS Physical Function, and subsequent corresponding anatomic-specific scales. The Brief Resilience Scale (BRS) was later added in the spring of 2020. A secure link to the surveys was sent to the patient’s email or phone prior to their appointment or accessed via an iPad at the clinic’s kiosk during their appointment. However, due to coronavirus cleaning protocols, forms were only sent to the patient’s email or phone beginning in March. Upon completion of July 2020 collection, we utilized STATA statistical software to analyze the data and identify trends in patient responses. For statistical analysis, we used chi-squared tests. Results: For the purpose of this study, we collected PROs from 7,557 patients between November 2019 and July 2020. After analyzing all data points in STATA, we determined differences in several collection categories, including the extent of completion and time of completion. Out of all patients prompted to complete one or more PROs, 37.20% of patients (n = 2811) did not complete any PROs, 11.86% of patients (n = 896) completed some but not all assigned PROs, and 50.95% of patients (n = 3850) completed all assigned PROs. Patients who completed all or some of their PROs completed them before their appointment 56.02% of the time (n = 2773), within an hour of their appointment 35.49% of the time (n = 1757), and after their appointment 8.48% of the time (n = 420) (see Figure 1). After separating PRO completion rates by month, we noticed a shift in compliance and patient response patterns before and during the coronavirus pandemic. From November to February, the average number of patients who completed one or more PROs was 80.20% (n = 1797) before dropping to an average of 57.74% (n = 5760; p <0.001) between March and July. The overall PRO compliance rate similarly dropped at our first PRO collection site, Clinic A, from an average of 86.75% to 76.81%, respectively (see Figure 2). Clinic operations were significantly reduced in late March due to the coronavirus pandemic, leading to majority telehealth visits and restrictions on clinic-wide shared iPad use. Though more than half of patients on average completed their PROs prior to their appointment throughout the year, this trend was especially prevalent during the coronavirus pandemic. For all or some PROs completed prior to appointment time, 42.34% were completed on average between November and February compared to 62.42% between March and July (see Figure 3). Beyond completion metrics, we observed a difference in appointment cancellations before and during the coronavirus pandemic. From November to February, 21.28% of patient appointments were canceled on average compared to 35.74% of appointments canceled on average from March to July. The number of cancelations during the coronavirus pandemic was significantly different than before the pandemic (p <0.001). Conclusions: In a short period of time (9 months) after initial implementation of a PRO platform, our practice has amassed significant data and reported outcomes for over 7,000 patients. Prior to the coronavirus pandemic, patients were relatively receptive to the new PRO collection platform with over 80% of patients completing some or all assigned PROs. However, following our clinic-wide cessation of shared iPads due to coronavirus concerns, we noted a significant decrease in our patient compliance (80.20% down to 57.74%; p <0.001). Though discontinuation of iPad and kiosk use likely only contributed in part to this significant decrease, this observation still suggests that shared devices may play an important role in promoting PRO compliance within the clinic. It’s also important to note the limitations in this study, including the staggered integration of each provider’s patients. This likely impacted the increased compliance rates in November with 282 patients versus 1677 patients in July. In addition, compliance may have been affected by the transition from two research staff members helping collect PROs to one research staff member. Methods of promoting compliance also changed as we shifted to telehealth visits. In March and April, members of our research team called patients about their incomplete PROs, likely prompting increased responses prior to patient appointments. We are utilizing these observations to improve our workflow for collecting patient PROs and increase compliance within the clinic. After reflecting on our initial findings, we feel the use of devices in clinic and full staff involvement can bolster completion of PROs and are important tools in order to best maintain a data collection platform with high patient compliance.

Software commonly consumes unexpectedly high amounts of memory, frequently due to programming idioms that are used to make software more reliable, maintainable and understandable. In the case of modern object-oriented systems this problem is partly due to creation of large numbers of co-existing isomorphic objects. Intuitively, two objects are isomorphic if they are of the same type, have identical values in corresponding primitive fields, and are such that corresponding reference fields themselves point to isomorphic objects. In other words, the portions of memory rooted at the two objects are isomorphic shape-wise as well as values-wise. A significant reduction in heap usage can therefore be achieved if the code is refactored to de-duplicate or share objects whenever possible instead of always creating distinct but possibly isomorphic objects. Such a refactoring, which employs a cache to keep track of objects created so far and to share them, is termed as object-sharing refactoring. In practice, object-sharing refactoring is commonly used, as it has the potential to reduce memory utilization significantly. However, manual refactoring is tedious and error prone. To support object-sharing refactoring we have (1) designed and implemented an approach to estimate memory-savings potential due to this refactoring, (2) espoused the idea of full initialization points of objects, and a static analysis to identify these points, and (3) proposed a scalable refinement-based points-to analysis for verifying the safety of object-sharing refactoring, but which has general applicability to several other verification tasks as well. 1) We present a dynamic analysis technique for estimating for all the allocation sites in a program, for a given input, the reduction in heap memory usage (in bytes, or as a percentage) to be obtained by employing object sharing at each site. The quantitative estimates produced by our technique of a user-observable benefit (i.e., actual memory savings) make it easier for developers to select sites to refactor. Experimentation with our estimation tool on real Java programs indicate that nearly all applications have potential for reduction of memory usage by object sharing, with a mean savings of 12.62% per benchmark. (2) We define a novel concept termed full-initialization points (FIPs) to characterize the points in the program where objects allocated at any chosen allocation site become fully initialized. We present a novel and conservative static analysis to detect FIPs for a given allocation site. By introducing code to cache and share objects at the FIPs suggested by our analysis, object-sharing refactoring was able to obtain a mean memory savings of 11.4% on a set of real Java benchmarks. (3) A standard points-to analysis approach, namely, the object sensitivity approach, uses an equal level of precision to represent symbolic objects allocated at all allocation sites in a program. This approach does not scale to large programs unless low levels of precision are used. We propose a novel, program-slicing based approach to improve the precision of object sensitivity analysis to answer user queries more precisely. Our slicing technique differs from most standard techniques for context-sensitive slicing of Java programs, which are based on system dependence graphs and which do not scale readily to larger programs. Our evaluation reveals that for a given time budget (6 hours per benchmark), our approach gives more precise results than the most precise results possible under the baseline object sensitivity analysis on nine of the 10 DaCapo benchmarks. Our approach exhibits 28% greater precision over the baseline object sensitivity approach in identifying allocation sites where object-sharing refactoring can be performed.

IMS is a standardized service architecture defined by 3GPP, ETSI, and IETF to provide multimedia services such as videoconferencing, VoD, and voice over IP. IMS is mainly based on the SIP protocol for session initialization. The convergence to full IP has advantages but also disadvantages. The latter are mainly inherited from the weaknesses of the IP protocol, in particular the QoS and the security aspects. It is in this context that this chapter is written. It has as main objective to analyze security in IMS networks as service layer in 4G to identify the most vulnerable points and propose security solutions that can be implemented without degrading the QoS.

Abstract Rotating instability (RI) in steam turbines is a phenomenon occurring during operation at very low volume flow conditions. Whereas RI is well-known in compressors, it is rather uncommon in turbines, where it is limited to the last stages of low-pressure steam turbines. The phenomenon has been studied numerically by means of viscous 3D CFD simulations employing mainly URANS equations. Given the possible difficulties to accurately predict heavily separated flows using such methods, this paper deals with the question whether the more sophisticated Improved Delayed Detached Eddy Simulation (iDDES) model is applicable in an industrial environment and whether it is capable of capturing the complex unsteady flow physics in a more realistic manner. For this purpose, the commercial CFD solver STAR-CCM+ is employed. A three-stage low-pressure model steam turbine featuring a non-axisymmetric inlet and an axial-radial diffuser is used as a test object. In order to capture the asymmetry, the model spans the full annulus and comprises the inlet section, all three stages, the diffuser as well as the exhaust hood. URANS and iDDES simulations have been performed at various low-volume flow part-load operating points and compared to test data. Unsteady pressure fluctuations at the casing as well as time-resolved probe traverse data have been used to validate the simulations. It is found that both models capture the overall flow physics well and that the iDDES model is superior at the most extreme part-load operating condition. In addition to the model accuracy and applicability of the CFD tool used, the paper discusses the challenges encountered during simulation setup as well as during initialization.

We focus on the simulation of shock-driven material mixing powered by flow instabilities dependent on initial conditions (IC) at the material interfaces. Beyond complex multi-scale resolution issues of shocks and variable density turbulence, we must address the equally difficult problem of predicting flow transition promoted by energy deposited at the interfacial layers during the shock-interface interactions. Transition involves IC-dependent, large-scale coherent-structure dynamics capturable by a large eddy simulation (LES) strategy, but not by unsteady Reynolds-Averaged Navier-Stokes (URANS) approaches based on equilibrium developed turbulence assumptions and single-point-closure modeling. On the engineering end of computations, reduced-dimensionality (1D/2D) versions of such URANS tend to be preferred for faster turnaround in full-scale configurations. With suitable initialization around each transition, URANS can be used to simulate the subsequent near-equilibrium weakly turbulent flow. We demonstrate 3D state-of-the-art URANS performance around one such (reshock) transition — in the context of a sequential LES/URANS strategy.