3D Gaussian Splatting (3DGS) [1] is a promising method for 3D reconstruction and novel view synthesis. However, training it with unconstrained images presents challenges due to transient objects that cause undesired floaters and ghosting artifacts. Although related works using Neural Radiance Fields (NeRF) [2] have attempted to address these issues, those techniques proved ineffective when directly applied to 3DGS. In this study, we propose an uncertainty estimation approach to assist 3DGS in reconstructing 3D scenes and removing transients. Specifically, we introduce a grouped uncertainty map using the Segment Anything Model [3] for per-area uncertainty estimation, combined with appearance embeddings to handle diverse lighting conditions. Experimental results on tourism photo collections [4] demonstrate that our method improves transient separation and rendering clarity. Furthermore, it facilitates effective color training and enables 3DGS to reconstruct target scenes from unconstrained images with fewer floaters or artifacts.

Modern electron-beam lithography (EBL) suffers from the long-range fogging effect that incurs undesired excessive exposure and thus layout pattern distortions. In this article, we propose an analytical placement algorithm to tackle the fogging effect. The underlying idea is to place standard cells, guided by our efficient, yet reasonably accurate fogging effect model, to minimize the fogging variation during placement, and thus the effect can be corrected by reducing dosage uniformly over the chip. We derive a fogging source modeling and further develop an efficient, accurate evaluation scheme to estimate the fogging effect by the fast Gauss transform with the Hermite expansion. The scheme achieves a 30.2X speedup over the traditional convolution computation, with only about 2.35% absolute average errors, which enables the iterative evaluation and variation minimization of the effect during global placement. We also develop fogging-aware legalization and detailed placement to further optimize the placement quality, while limiting fogging variation. The experimental results show that our algorithm can effectively reduce the fogging variation by 35.4%, while maintaining high wirelength quality, at reasonable runtime.

Transfer learning and Transformer-based language models play important roles in modern natural language processing research community. In this paper, we propose Transformer model's fine-tuning and data augmentation (TMFTDA) techniques for conversational texts and noisy user-generated content. We use two NTCIR-15 tasks, namely the first Dialogue Evaluation (DialEval-1) task and the second Numeral Attachment in Financial Tweets (FinNum-2) task, to evaluate the efficacy of TMFTDA. Experimental results show that TMFTDA substantially outperforms the baselines model of Bidirectional Long Short-Term Memory (Bi-LSTM) in multi-turn dialogue system evaluation at DialEval-1's Dialogue Quality (DQ) and Nugget Detection (ND) subtasks. Moreover, TMFTDA performs to a satisfactory level at FinNum-2 with a model of Cross-lingual Language Models using a Robustly Optimized BERT Pretraining Approach (XLM-RoBERTa). The research contribution of this paper is that, we help shed some light on the usefulness of TMFTDA, for conversational texts and noisy user-generated content in social media text analytics.

Modern electron beam lithography (EBL) suffers from the long-range fogging effect which incurs undesired excessive exposure and thus layout pattern distortions. In this paper, we propose the first placement algorithm to tackle the fogging effect. The underlying idea is to place standard cells, guided by our efficient, yet reasonably accurate fogging effect model to minimize the fogging variation during placement, and thus the effect can be corrected by reducing dosage uniformly over the chip. We use fast Gauss transform with Hermite expansion for convolution approximation to estimate the effect. This approximation achieves a 26.5X speedup over traditional convolution computation, with only about 0.33% absolute average errors. We derive a fogging source model and further develop an efficient, accurate evaluation scheme to estimate the fogging effect by fast Gauss transform. The scheme achieves a 30.2X speedup over traditional convolution computation, with only about 2.35% absolute average errors, which enables the iterative evaluation and variation minimization of the effect during analytical global placement. We also develop fogging-aware legalization and detailed placement to further optimize the placement quality, while maintaining fogging variation. Experimental results show that our algorithm can effectively reduce the fogging variation by 15.5%, while maintaining high wirelength quality, at reasonable runtime.

As nanometer technology advances, conventional optical proximity correction (OPC) that minimizes the edge placement error (EPE) at the nominal process condition alone often leads to poor process windows. To improve the mask printability across various process corners, process-window OPC optimizes EPE for multiple process corners, but often suffers long runtime, due to repeated lithographic simulations. This paper presents an efficient process variation (PV)-aware mask optimization framework, namely PVOPC, to simultaneously minimize EPE and PV band with fast convergence. The PVOPC framework includes EPE-sensitivity-driven dynamic fragmentation, PV-aware EPE modeling, and correction with three new EPE-converging techniques and a systematic subresolution-assisted feature insertion algorithm. Experimental results show that our approach efficiently achieves high-quality EPE and PV band results.

As nanometer technology advances, conventional OPC (Optical Proximity Correction) that minimizes the EPE (Edge Placement Error) at the nominal corner alone often leads to poor process window. To improve the mask printability across various process corners, process-window OPC optimizes EPE for multiple process corners, but often suffers long runtime, due to repeated lithographic simulations. This paper presents an efficient process-variation-aware mask optimization framework, namely PVOPC (Process-Variation OPC), to simultaneously minimize EPE and PV (Process-Variation) band with fast convergence. The PVOPC framework includes EPE-sensitivity-driven dynamic fragmentation, process-variation-aware EPE modeling, and post correction with three new EPE-converging techniques and a systematic sub-resolution assisted feature insertion algorithm. Experimental results show that our approach efficiently achieves high-quality EPE and PV band results.

Measurement of end-to-end available bandwidth has received considerable attention due to its potential use in improving QoS. Available bandwidth enables the sending rate to adapt to network conditions, so that packet loss, caused by congestion, can be significantly reduced before error control mechanisms are finally employed. To this end, we propose a probing noise resilient available bandwidth estimation scheme, called JitterPath, which is adaptive to both the fluid and bursty traffic models. Two key factors, one-way delay jitter and accumulated queuing delay, are both exploited to predict the type of queuing region for each packet pair. Then, the bottleneck utilization information included in the joint queuing regions is estimated and used to quantify the captured traffic ratio, which indicates the relationship between the probing rate and available bandwidth. The contributions of our method are as follows: 1) JitterPath can work without being restricted to fluid traffic models; 2) since JitterPath does not directly use the bottleneck link capacity to calculate the available bandwidth, it is feasible for use in a multihop environment with a single bottleneck; and 3) JitterPath inherently reduces the impact of probing noises under the bursty cross traffic model. Extensive simulations, Internet experiments, and comparisons with other methods were conducted to verify the effectiveness of our method under both single-hop and multihop environments

In this paper, we present a real-time multi-objects tracking system which can detect various types of moving objects in image sequences of traffic video obtained from a stationary video camera. Using the adaptive background reconstruction technique can effectively handle with environmental changes and obtain good results of objects extraction. Besides, we introduce a robust region-and feature-based tracking algorithm with plentiful features to track correct objects continuously. After tracking objects successfully, we can analyze the tracked objects' properties and recognize their behavior for extracting some useful traffic parameters. According to the structure of our proposed algorithms, we implemented a tracking system including the functions of objects classification and accident prediction. Experiments were conducted on real-life traffic video of some intersection and testing datasets of other surveillance research. The results proved the algorithms we proposed achieved robust segmentation of moving objects and successful tracking with objects occlusion or splitting events. The implemented system also extracted useful traffic parameters.

We propose a one-way delay jitter based scheme, "jitterpath," for available bandwidth estimation. Common assumptions, including use of the fluid traffic model and use of the bottleneck link capacity, that have been made in the literature are relaxed in this study. We exploit one-way delay jitter and accumulated queuing delay to predict the type of a queuing region for each packet pair. In addition, we quantify the captured traffic ratio, which is defined as the total output gaps of joint queuing regions per total input gaps, and use it to derive the relationship between probing rate and available bandwidth. We further investigate how the estimation resolution and the probing noise ratio are related to the accuracy of available bandwidth estimation. Extensive simulations and real-network experiment have been conducted and comparisons with other methods have been made to verify the effectiveness of our method, no matter whether single-hop or multi-hop environments are considered

Video transmission needs a stable sending rate in order that the video can be displayed showing uniform quality. In addition, a lower packet loss rate is rather helpful in reducing video quality degradation. Therefore, reliable available bandwidth estimation becomes an indispensable step towards robust transmission of multimedia data. The existing available bandwidth estimation methods cannot deal with the false estimation problem and thereby precise estimation is impossible. In this paper, we propose a reliable available bandwidth estimation method, based on distinguishing queuing regions and resolving false estimations. Promising simulation results indicate that our method can obtain the available bandwidth precisely, no matter what the network environment is; single-bottleneck or multiple-bottleneck.