Parameter-efficient fine-tuning of pre-trained multilingual speech models can significantly enhance the recognition performance target languages. However, traditional parameter-efficient methods, such as adapter tuning, often face challenges related to random initialization. This lead suboptimal when adapting languages with limited resources. To address this issue, paper introduces TAML-Adapter, which utilizes Task-Agnostic Meta-Learning algorithm initialize parameters adapters before in...

Abstract Limited data availability remains a significant challenge for Whisper’s low-resource speech recognition performance, falling short of practical application requirements. While previous studies have successfully reduced the error rates target language through fine-tuning, comprehensive exploration and analysis fine-tuning capabilities advantages disadvantages various strategies are still lacking. This paper aims to fill this gap by conducting experimental performance using five with...

Data centers, the critical infrastructure underpinning Cloud computing, often employ Software-Defined Networks (SDN) to manage cluster, wide-area and enterprise networks. As network forwarding in SDN is dynamically programmed by controllers, it crucial ensure that controller intent correctly translated into underlying rules. Therefore, detecting locating anomalies a fundamental problem production Existing research proposals, roughly categorized probing-based, packet piggybacking-based, flow...

Multi-domain network resource reservation systems are being deployed, driven by the demand and substantial benefits of providing predictable resources. However, a major lack existing is their coarse granularity, due to participating networks' concern revealing sensitive information, which can result in inefficiencies. This paper presents Mercator, novel multi-domain discovery system provide fine-grained, global for collaborative sciences. The foundation Mercator abstraction through...

Image segmentation is an important research in image processing and machine vision which automated driving can be seen the main application scene of algorithms. Due to many constraints power supply communication in-vehicle systems, vast majority current algorithms are implemented based on deep learning model. Despite ultrahigh accuracy, problem mesh artifacts being too severe obvious, high cost, computational, consumption devices required difficult apply real-world scenarios. It focus this...

Abstract Camera calibration, image feature detection, matching and other aspects have become barriers that traditional 3D reconstruction methods are difficult to break through. The important role of deep learning in data detection classification has a impact on the real world, research hotspot at home abroad deal with this problem. In paper, method sequence based depth is proposed. Firstly, principle introduced. Then, new studied discussed combination theory. Finally, conclusion prospect given.

Holographic stereogram comprises a hotspot in the field of three-dimensional (3D) display. It can reconstruct light information real and virtual scenes at same time, further improving comprehensibility scene achieving “augmentation” scene. In this paper, an augmented reality-holographic based on 3D reconstruction is proposed. First, point cloud data generated by VisualSFM software, then mesh model reconstructed MeshLab software. The obtained are rendered simultaneously to obtain fusion...

Recently, feature relation learning has drawn widespread attention in cross-spectral image patch matching. However, existing related research focuses on extracting diverse relations between features and ignores sufficient intrinsic representations of individual patches. Therefore, an innovative relational representation idea is proposed for the first time, which simultaneously sufficiently mining patches features. Based this, we construct a lightweight Relational Representation Learning...

Recently, cross-spectral image patch matching based on feature relation learning has attracted extensive attention. However, performance bottleneck problems have gradually emerged in existing methods. To address this challenge, we make the first attempt to explore a stable and efficient bridge between descriptor metric learning, construct knowledge-guided network (KGL-Net), which achieves amazing improvements while abandoning complex structures. Specifically, find that there is extraction...

Abstract To reduce the view-flipping effect and enhance viewing resolution, modulation characteristics of hogel based holographic stereogram is constructed validated. The performance analyzed, results indicate that decreasing size beneficial to reduction view flipping, however, which will result in significant diffraction can degrade reconstruction quality. Furthermore, a diffraction-limited imaging model established, where both limited aperture defocused aberration object point are...

Aiming at the problems of missing important features, inconspicuous details and unclear textures in fusion multimodal medical images, this paper proposes a method computed tomography (CT) image magnetic resonance imaging (MRI) using generative adversarial network (GAN) convolutional neural (CNN) under enhancement. The generator aimed high-frequency feature images used double discriminators to target after inverse transform; Then were fused by trained GAN model, low-frequency CNN pre-training...

In this paper, an optical field coding method for the fusion of real and virtual scenes is proposed to implement augmented reality (AR)-based holographic stereogram. The occlusion relationship between analyzed, a strategy based on instance segmentation depth determination proposed. A three-dimensional (3D) scene sampling system built, foreground contour sampled perspective image extracted by Mask R-CNN algorithm. 3D rendered computer obtain images as well their maps. According relation...

We investigate how the splicing mode of a holographic element (hogel) affects reconstruction 3D scene to improve resolution stereogram fabricated using effective perspective image segmentation and mosaicking method (EPISM). First, effect hogel spatial multiplexing on recording is studied based mechanism interference fringes in medium. Second, combined with influence multiple exposures hologram’s diffraction efficiency, efficiency analyzed mode. The then regarded as special optical imaging...

Different light-fields information collection methods will cause different cannot be fused and recorded into the holographic stereogram. In order to print stereogram recording of virtualreal fusion, depth-image-based rendering (DIBR) technology is applied fabricated by effective perspective images' segmentation mosaicking method (EPISM). This firstly obtains matching real scene sampling parameters according virtual scene, combines with improved DIBR draw initial images without holes or crack...

In this paper, the principle as well implementation of EPISM method are introduced firstly. order to evaluate reconstruction quality better, imaging process based holographic stereogram is regarded a general optical system imaging, and modeling optimization proposed from two different aspects angular spectrum spatial domain. analysis angle theory, exit pupil function model simplified firstly transfer (OTF) with defocusing aberrations was established. domain analysis, modulation...

In this paper, the light field regulation characteristics of size and transmittance function hogel in holographic stereogram are analyzed, which shows that display quality can be effectively improved by using special‐shaped pupil. The limitation optimizing is analyzed from perspective resolution view‐flipping effect. By establishing diffraction limited imaging model with aberration stereogram, point spread (PSF) characterizing image propagation obtained, important variable affecting...

In this paper, a perspective image fusion algorithm of real-world and artificial scenes based on depth maps is proposed applied to holographic printing realize augmented reality display stereograms. The basic principle the introduced, with correct occlusion relation implemented theoretically. Under condition sampling only once, an generation adjustable implemented. We use effective segmentation mosaicking method for optical experiments, verify effectiveness method.

Holographic stereogram opens up a new way for holographic 3D display of objects, and has high research value significance in commercial, military other aspects. In order to write real virtual three-dimensional (3D) scenes at the same time, so as achieve effect augmented reality, light field fusion method is proposed. The basic principle introduced. sampling introduced completed, depth image based rendering (DIBR) algorithm applied regularization densification sampled images. projection...