Jie Zhu, Junhui Li, Muhua Longhua Qian, Min Zhang, Guodong Zhou. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint (EMNLP-IJCNLP). 2019.

This article proposes a network, referred to as Multi-View Stereo TRansformer (MVSTR) for depth estimation from multi-view images. By modeling long-range dependencies and epipolar geometry, the proposed MVSTR is capable of extracting dense features with global context 3D consistency, which are crucial reliable matching in stereo (MVS). Specifically, tackle problem limited receptive field existing CNN-based MVS methods, global-context Transformer module designed establish intra-view so that...

Recent years have witnessed the remarkable success of vision-language model in various computer vision tasks. However, how to exploit semantic language knowledge advance real-world stereoscopic image super-resolution remains a challenging problem. This paper proposes model-based (VLM-SSR) method, which CLIP is exploited facilitate SR training-free manner. Specifically, by designing visual prompts for infer region similarity, prompt-guided information aggregation mechanism presented capture...

This paper proposes a network, referred to as MVSTR, for Multi-View Stereo (MVS). It is built upon Transformer and capable of extracting dense features with global context 3D consistency, which are crucial achieving reliable matching MVS. Specifically, tackle the problem limited receptive field existing CNN-based MVS methods, global-context module first proposed explore intra-view context. In addition, further enable be 3D-consistent, 3D-geometry well-designed cross-view attention mechanism...

Most existing learning-based methods for stereoscopic image super-resolution rely on a great number of high-resolution images as labels. To alleviate the problem data dependency, this paper proposes self-supervised pretraining-based method (SelfSSR). Specifically, to develop pretext task images, parallax-aware masking strategy (PAMS) is designed adaptively mask matching areas left and right views. With PAMS, network encouraged effectively predict missing information input images. Besides,...

Recent studies on AMR-to-text generation often formalize the task as a sequence-to-sequence (seq2seq) learning problem by converting an Abstract Meaning Representation (AMR) graph into word sequence. Graph structures are further modeled seq2seq framework in order to utilize structural information AMR graphs. However, previous approaches only consider relations between directly connected concepts while ignoring rich structure In this paper we eliminate such strong limitation and propose novel...

Abstract Recently, deep learning‐based stereo matching has achieved great success. However, models trained on the source domain dataset encounter substantial performance degradation when directly tested an unseen target because of neglecting generalization to out‐of‐distribution (OOD) samples. This paper proposes a sample generation‐based network (SGDG‐Net) for matching. Specifically, expand distribution span training samples, OOD samples are generated assist training. To effectively...

Traditional methods of dimension matching has some disadvantages large task computing and discarding useful data in Radar network fusion. So they can't meet the needs air intelligence fusion anymore. The fractal is a new way to solve radar measurement different scales without matching. This method bases on celestial coordinate system, which can be seen as two parts, Centrifugal distance. It get system track based Celestial with process. Simulation provided show that higher accuracy less this paper.

Objective: To investigate the efficacy and safety of CT-guided radiofrequency ablation posterior root spinal nerve in treatment postherpetic neuralgia (PHN). Methods: A total 102 PHN patients (42 males 60 females) aged (69.7±9.4) years who underwent Department Pain Medicine Affiliated Hospital Jiaxing University from January 2017 to April 2020 were retrospectively included. Patients followed up, numerical rating scale (NRS) score, Pittsburgh sleep quality index (PSQI), satisfaction score...

Abstract Due to America’s leadership in the photography industry, American wars have been repeatedly brought screen. The photographer’s attitude of excellence this kind has also forged classics film industry. After contemporary era, with advent consumer and picture-reading a clearer narrative strategy emerged cinema that differs from earlier adaptations war literature. In order explore aesthetic implications ethical values inherent photography, as well correct some previous biases, we aim...

We present a data-driven approach that colorizes 3D furniture models and indoor scenes by leveraging images on the internet. Our is able to colorize automatically according an example image. The core learn image-guided mesh segmentation segment model into different parts image object. Given scene, system supports colorization-by-example, has ability recommend colorization scheme consistent with user-desired color theme. latter realized formulating problem as Markov random field imposes user...

Abstract Phoxim is a significant insecticide, and its current synthesis method has some shortcomings such as the high risk of explosion trans structure (main impurity) hard to control. Our work solved above disadvantages by introducing macromolecular alcohol (benzyl alcohol, etc.) starting material optimizing intermediate reaction conditions. Compared with route, synthetic following advantages: (1) benzyl nitrite boiling point strong safety; (2) α -cyanobenzaldehyde oxime sodium are almost...