Almost all previous deep learning-based multi-view stereo (MVS) approaches focus on improving reconstruction quality. Besides quality, efficiency is also a desirable feature for MVS in real scenarios. Towards this end, paper presents Fast-MVSNet, novel sparse-to-dense coarse-to-fine framework, fast and accurate depth estimation MVS. Specifically, our we first construct sparse cost volume learning high-resolution map. Then leverage small-scale convolutional neural network to encode the...

How should we integrate representations from complementary sensors for autonomous driving? Geometry-based fusion has shown promise perception (e.g., object detection, motion forecasting). However, in the context of end-to-end driving, find that imitation learning based on existing sensor methods underperforms complex driving scenarios with a high density dynamic agents. Therefore, propose TransFuser, mechanism to image and LiDAR using self-attention. Our approach uses transformer modules at...

In recent years, neural implicit surface reconstruction methods have become popular for multi-view 3D reconstruction. contrast to traditional stereo methods, these approaches tend produce smoother and more complete reconstructions due the inductive smoothness bias of networks. State-of-the-art allow high-quality simple scenes from many input views. Yet, their performance drops significantly larger complex captured sparse viewpoints. This is caused primarily by inherent ambiguity in RGB loss...

3D Gaussian Splatting (3DGS) has recently revolutionized radiance field reconstruction, achieving high quality novel view synthesis and fast rendering speed. However, 3DGS fails to accurately represent surfaces due the multi-view inconsistent nature of Gaussians. We present 2D (2DGS), a approach model reconstruct geometrically accurate fields from images. Our key idea is collapse volume into set oriented planar disks. Unlike Gaussians, Gaussians provide view-consistent geometry while...

Single-image piece-wise planar 3D reconstruction aims to simultaneously segment plane instances and recover parameters from an image. Most recent approaches leverage convolutional neural networks (CNNs) achieve promising results. However, these methods are limited detecting a fixed number of planes with certain learned order. To tackle this problem, we propose novel two-stage method based on associative embedding, inspired by its success in instance segmentation. In the first stage, train...

An Axial Shifted MLP architecture (AS-MLP) is proposed in this paper. Different from MLP-Mixer, where the global spatial feature encoded for information flow through matrix transposition and one token-mixing MLP, we pay more attention to local features interaction. By axially shifting channels of map, AS-MLP able obtain different axial directions, which captures dependencies. Such an operation enables us utilize a pure achieve same receptive field as CNN-like architecture. We can also design...

Multiple near frontal-parallel planes based depth representation demonstrated impressive results in self-supervised monocular estimation (MDE). Whereas, such a would cause the discontinuity of ground as it is perpendicular to planes, which detrimental identification drivable space autonomous driving. In this paper, we propose PlaneDepth, novel orthogonal presentation, including vertical and planes. PlaneDepth estimates distribution using Laplacian Mixture Model on for an input image. These...

Recently, 3D Gaussian Splatting (3DGS) has demonstrated impressive novel view synthesis results, while allowing the rendering of high-resolution images in real-time. However, leveraging Gaussians for surface reconstruction poses significant challenges due to explicit and disconnected nature Gaussians. In this work, we present Opacity Fields (GOF), a approach efficient, high-quality, adaptive unbounded scenes. Our GOF is derived from ray-tracing-based volume Gaussians, enabling direct...

Abstract As a unique Cairo pentagonal 2D material, palladium phosphide sulfide (PdPS) has garnered immense interests due to its excellent optoelectronic properties, anisotropic electronic transport behavior, and good air‐stability. In addition, puckered pentagon structure renders an ultralow thermal conductivity, making it promising candidate for thermoelectrics applications. However, thermoelectric not been studied until now challenges in obtaining the atomic thin PdPS flake further...

This study aims to investigate the association between serum copper (Cu), selenium (Se), zinc (Zn), Se/Cu and Zn/Cu ratios risk of sarcopenia. In this study, which involved 2766 adults aged ≥ 20 years enrolled in National Health Nutrition Examination Survey (NHANES) from 2011 2016, multivariable logistic regression, restricted cubic spline (RCS) models mediation analyses were used. After full adjustment, regression revealed that higher levels correlated with an increased Conversely, (OR...

Small‐molecule organic carbonyl compounds (SMOCMs) featuring high theoretical capacities are promising cathodes for lithium‐ion batteries (LIBs), but facing challenges in cycling stability and rate performance owing to their solubility electrolytes low conductivity. Herein, we propose a novel architecture wherein nanosized SMOCMs with N‐heterocycle‐extended π‐conjugation uniformly immobilized on reduced graphene oxide (rGO). This approach leverages the N‐heterocycles create additional active...

Small‐molecule organic carbonyl compounds (SMOCMs) featuring high theoretical capacities are promising cathodes for lithium‐ion batteries (LIBs), but facing challenges in cycling stability and rate performance owing to their solubility electrolytes low conductivity. Herein, we propose a novel architecture wherein nanosized SMOCMs with N‐heterocycle‐extended π‐conjugation uniformly immobilized on reduced graphene oxide (rGO). This approach leverages the N‐heterocycles create additional active...

Abstract Malignant bone tumor defects are difficult to treat because of the simultaneous need for treatment and bone‐repair promotion. This study presents a bioactive composite scaffold (T‐rGO@Au) personalized defect repair treatment. The T‐rGO@Au has porous structure, its mechanical properties close those human cancellous bone. can induce upregulation osteopontin (OPN), RUNX‐2, osteocalcin (OCN) gene expression. In vivo experiments showed that volume/total volume (BV/TV) ratio with was...

Abstract The continuous miniaturization of 2D electronic circuits results in increased power density during device operation, leading to heat localization and placing higher demands on their performance thresholds. risk thermal breakdown subsequent damage, due the energy dissipation semiconductor field‐effect transistors (FETs) supported bulk substrates, represents a significant challenge maintaining optimal performance. Herein, this study investigates behavior multilayer PdSe 2 FETs for...

Diabetic retinopathy (DR), as a leading ocular disease, is often with complication of diabetic macular edema (DME). However, most existing works only aim at DR grading but ignore the DME diagnosis, doctors will do both tasks simultaneously. In this paper, motivated by advantages multi-task learning for image classification, and to mimic behavior clinicians in visual inspection patients, we propose feature Separation Union Network (SUNet) simultaneous grading. Further, improve...

We consider a class of angular Gaussian distributions that allows different degrees isotropy for directional random variables arbitrary dimension. To incorporate constraints imposed on the original model parameters, we propose new parameterization distribution so all parameters are free constraints. Via parameterization, translate problem maximum likelihood estimation subject to complex routine optimization constraints, which in turn leads theoretically sound and numerically stable...