Recent progress in diffusion models has revolutionized the popular technology of text-to-image generation. While existing approaches could produce photorealistic high-resolution images with text conditions, there are still several open problems to be solved, which limits further improvement image fidelity and relevancy. In this paper, we propose ERNIE-ViLG 2.0, a large-scale Chinese model, progressively upgrade quality generated by: (1) incorporating fine-grained textual visual knowledge key...

Knowing the reasoning chains from knowledge to predicted answers can help construct an explainable question answering (QA) system. Advances on QA explanation propose explain with entailment trees composed of multiple steps. While current work proposes generate end-to-end generative models, steps in generated are not constrained and could be unreliable. In this paper, we METGEN, a Module-based Entailment Tree GENeration framework that has modules controller. Given several supporting...

Xintong Yu, Hongming Zhang, Yangqiu Song, Yan Changshui Zhang. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint (EMNLP-IJCNLP). 2019.

This study has designed and developed a smart data glove based on five-channel flexible capacitive stretch sensors six-axis inertial measurement unit (IMU) to recognize 25 static hand gestures ten dynamic for amphibious communication. The are fabricated capture finger motion in order integrated with IMU gestures. also proposes novel hierarchical gesture recognition (AHGR) model. model can adaptively switch between large complex lightweight models environmental changes ensure accuracy...

Halide perovskites have attracted increasingly attention as “rising star” materials for advanced photonics and optoelectronics. Construction micro‐/nano‐architecture of will provide a good platform to investigate optimize the fundamental photon–matter–structure interaction. It also improve properties, pixelate miniaturize integration versatile optoelectronic devices emerging applications. In this regard, femtosecond (fs) laser processing technique has been widely used fabricate with high...

Abstract Fine geological modeling leads to accurate reservoirs numerical simulations. Fractured biogenic limestone has abundant storage spaces and flow paths accumulate oil gas. The complexity diversity of fractured also lead challenges in accurately characterizing its pore volume remaining oil. This investigation aimed enhance the understanding biotite reservoir properties via modeling. Numerical simulations were used characterize during late stage field development. Considering differences...

Top-k performance has recently received increasing attention in large data categories. Advances, like a top-k multiclass support vector machine (SVM), have consistently improved the accuracy. However, key ingredient state-of-the-art optimization scheme based upon stochastic dual coordinate ascent relies on sorting method, which yields O(d log d) complexity. In this paper, we leverage semismoothness of problem and propose an optimized SVM algorithm, employs semismooth Newton algorithm for...

Grounding a pronoun to visual object it refers requires complex reasoning from various information sources, especially in conversational scenarios. For example, when people conversation talk about something all speakers can see, they often directly use pronouns (e.g., it) refer without previous introduction. This fact brings huge challenge for modern natural language understanding systems, particularly conventional context-based coreference models. To tackle this challenge, paper, we...

With the rapid development of computer vision and artificial intelligence, human-computer interaction has become an inevitable part people's lives. Gestures can bring more natural, comfortable, effective communication between people machines. However, in some complex scenarios, such as rooms with looming lighting, robustness universality hand gesture recognition based on traditional cameras are insufficient, supporting algorithms tend to underperform real-time, especially for embedded...

Abstract In accordance with the problem of worst at convergence BP Neural Network, elevator fault-prediction-model is proposed based on Improved PSO-BP Algorithm in this paper. The method uses mathematical operation mechanism to analyze characteristic be studied. Then prediction model fault established. This paper tests same data by using different models. experimental results show that has higher accuracy and convergence. provides a reliability.

Abstract Aiming at the problem that current elevator monitoring system cannot detect accidental fall of passengers, this paper proposes a detection method based on machine vision and multi-feature fusion. First, moving targets were extracted by ViBe algorithm, then human body was marked with an external rectangle. Three characteristic parameters, namely aspect ratio, effective area ratio centroid acceleration body, calculated. At last, thresholds set SVM classification training conducted to...

This paper presents an enhanced ground vehicle localization method designed to address the challenges associated with state estimation for autonomous vehicles operating in diverse environments. The focus is specifically on precise of position and orientation both local global coordinate systems. proposed approach integrates estimates generated by existing visual-inertial odometry (VIO) methods into information obtained from Global Navigation Satellite System (GNSS). integration achieved...

Multiple sclerosis (MS) is a condition that affects the veins and small blood vessels. Previous research suggests individuals with MS have an increased risk of vascular events higher mortality rates. However, relationship between cerebral vessel disease (CSVD) remains uncertain. This study aims to investigate association lacunes. A prospective observational was conducted, including total 112 participants, which 46 had 66 CSVD. All participants underwent MRI scan battery neurological...

Abstract The function of traditional elevator monitoring system is relatively simple. It cannot realize on-demand maintenance and fault warning. To solve such problems, this paper builds an intelligent Internet Things based on multi-sensor information fusion, bus communication probability statistical analysis technology. Managers can access the control platform through smart terminals or web browsers to achieve data query, video monitoring, alarm management system, health other functions,...

Existing fully supervised event extraction models achieve advanced performance with large-scale labeled data. However, when new types emerge and annotations are scarce, it is hard for the to master limited annotations. In contrast, humans can learn understand only a few examples in guideline. this paper, we work on challenging yet more realistic setting, few-example extraction. It requires sentences guidelines as training data, so that do not need collect each time emerge. As tend overfit...

We propose a foreground segmentation algorithm that does extraction under different scales and refines the result by matting. First, input image is filtered resampled to 5 resolutions. Then each of them segmented adaptive figure-ground classification best automatically selected an evaluation score maximizes difference between background. This upsampled original size, corresponding trimap built. Closed-form matting employed label boundary region, refined final classification. Experiments show...

The diversity and versatility of display devices today imposes new demands on digital elevation models (DEMs). This paper proposes a resizing method for 3D visualization DEMs based topographic feature. proposed improves seaming carving algorithm to resize instead images according the characterristics DEMs. considers not only geometric constraints but also characteristics Being different from traditional reduction expansion methods, resizes DEMs, preserves Digital Elevation Model. is...

Abstract In this paper, the research object is ternary lithium battery with NCA as positive material and graphite negative material. The model of one-dimensional capacity fade established by COMSOL. At different simulated temperatures, changes SOC voltage drop SEI film on surface were analyzed. results show that average decreases increase temperature impedance.

Resolving pronouns to their referents has long been studied as a fundamental natural language understanding problem. Previous works on pronoun coreference resolution (PCR) mostly focus resolving mentions in text while ignoring the exophoric scenario. Exophoric are common daily communications, where speakers may directly use refer some objects present environment without introducing first. Although such not mentioned dialogue text, they can often be disambiguated by general topics of...