Modeling relation between actors is important for recognizing group activity in a multi-person scene. This paper aims at learning discriminative efficiently using deep models. To this end, we propose to build flexible and efficient Actor Relation Graph (ARG) simultaneously capture the appearance position actors. Thanks Convolutional Network, connections ARG could be automatically learned from videos an end-to-end manner, inference on performed with standard matrix operations. Furthermore,...

Deep learning (DL)-based object detection algorithms have gained impressive achievements in natural images and gradually matured recent years. However, compared with images, remote sensing are faced severe challenges due to the complex backgrounds difficult of small objects dense scenes. To address these problems, a novel one-stage model named MDCT is proposed based on multi-kernel dilated convolution (MDC) block transformer block. Firstly, new feature enhancement module, MDC block,...

This paper addresses the task of estimating spatially-varying reflectance (i.e., SVBRDF) from a single, casually captured image. Central to our method is highlight-aware (HA) convolution operation and two-stream neural network equipped with proper training losses. Our HA convolution, as novel variant standard (ST) directly modulates kernels under guidance automatically learned masks representing potentially overexposed highlight regions. It helps reduce impact strong specular highlights on...

Fitting primitives for point cloud data to obtain a structural representation has been widely adopted reverse engineering and other graphics applications. Existing segmentation-based approaches only segment primitive patches but ignore edges that indicate boundaries of primitives, leading inaccurate incomplete reconstruction. To fill the gap, we present novel surface edge detection network (SED-Net) accurate geometric fitting clouds. The key idea is learn parametric surfaces (including...

Depth estimation is a fundamental task in many vision applications. With the popularity of omnidirectional cameras, it becomes new trend to tackle this problem spherical space. In paper, we propose learning-based method for predicting dense depth values scene from monocular image. An image has full field-of-view, providing much more complete descriptions than perspective images. However, fully-convolutional networks that most current solutions rely on fail capture rich global contexts...

In the field of Moving Infrared Small Target Detection (MIRSTD), current methods typically use sequential modeling with two individual modules for spatial and temporal processing. However, such a strategy lacks clear guidance on motion displacement difference between moving targets background noise, thereby limiting feature discriminability resulting in error-prone target localization. This paper addresses this issue from clip frame levels proposes novel architecture MOCID MIRSTD. For...

Abstract Effective compression of densely sampled BRDF measurements is critical for many graphical or vision applications. In this paper, we present DeepBRDF, a deep‐learning‐based representation that can significantly reduce the dimensionality measured BRDFs while enjoying high quality recovery. We consider each as sequence image slices and design deep autoencoder with masked L 2 loss to discover nonlinear low‐dimensional latent space high‐dimensional input data. Thorough experiments verify...

Reconstructing objects from posed images is a crucial and complex task in computer graphics vision. While NeRF-based neural reconstruction methods have exhibited impressive ability, they tend to be time-comsuming. Recent strategies adopted 3D Gaussian Splatting (3D-GS) for inverse rendering, which led quick effective outcomes. However, these techniques generally difficulty producing believable geometries materials glossy objects, challenge that stems the inherent ambiguities of rendering. To...

Parametric edge reconstruction for point cloud data is a fundamental problem in computer graphics. Existing methods first classify points as either (including corners) or non-edge points, and then fit parametric edges to the points. However, few are exactly sampled on practical scenarios, leading significant fitting errors reconstructed edges. Prominent deep learning-based also primarily emphasize overlooking potential of areas. Given that sparse non-uniform cannot provide adequate...

Lane detection based on computer vision is a key technology of Automatic Drive System for intelligent vehicles. In this paper, we propose real-time and efficient lane algorithm that can detect lanes appearing in urban streets highway roads under complex background. order to enhance boundary information be suitable various light conditions, adopt canny edge get good feature points. We use the generalized curve parameter model, which describe both straight curved lanes. an improved random...

Existing convolutional neural networks have achieved great success in recovering Spatially Varying Bidirectional Surface Reflectance Distribution Function (SVBRDF) maps from a single image. However, they mainly focus on handling low-resolution (e.g., 256 × 256) inputs. Ultra-High Resolution (UHR) material are notoriously difficult to acquire by existing because (1) finite computational resources set bounds for input receptive fields and output resolutions, (2) layers operate locally lack the...

Unsupervised completion of real scene objects is vital importance but still remains extremely challenging in preserving input shapes, predicting accurate results, and adapting to multi-category data. To solve these problems, we propose this paper an Symmetric Shape-Preserving Autoencoding Network, termed USSPA, predict complete point clouds from scenes. One our main observations that many natural manmade exhibit significant symmetries. accommodate this, devise a symmetry learning module...

Current available antidepressants exhibit low remission rate with a long response lag time. Growing evidence has demonstrated acute sub-anesthetic dose of ketamine exerts rapid, robust, and lasting antidepressant effects. However, term use tends to elicit its adverse reactions. The present study aimed investigate the antidepressant-like effects intermittent consecutive administrations on chronic unpredictable mild stress (CUMS) rats, determine whether can redeem time for treatment classic...

Image stitching for two images without a global transformation between them is notoriously difficult. In this paper, noticing the importance of semantic planar structures under perspective geometry, we propose new image method which stitches by allowing alignment set matched dominant regions. Clearly different from previous methods resorting to plane segmentation, key our approach utilize rich information directly RGB extract regions with deep Convolutional Neural Network (CNN). We...

A simple method was proposed to control the nanotribology behaviors of monocrystalline silicon against SiO2 microsphere by adjusting relative humidity (RH). Experimental results indicated that adhesion work, friction coefficient, and nanowear significantly varied between 60% 90% RH. Under RH, work 119 mN/m, coefficient about 0.53. However, decreased ∼70 mN/m ∼0.3 under respectively. An apparent wear track ∼13 nm deep formed on surface whereas no obvious scar observed Analysis such...

Abstract In this paper, a novel planar microwave retroreflector based on transmissive gradient metasurface combined with curved metal mirror is proposed and demonstrated. The can efficiently converge wide-angle incident wave to pre-designed behind it proper distance, which acts as an effective reflective surface that greatly enhance the backscattering of view. According full-wave simulations, perform excellent retroreflective effect for microwaves angle view between −30° 30° range. A...

Monte Carlo (MC) methods for light transport simulation are flexible and general but typically suffer from high variance slow convergence. Gradientdomain rendering alleviates this problem by additionally generating image gradients reformulating as a screened Poisson reconstruction problem. To improve the quality performance of reconstruction, we propose novel practical deep learning based approach in paper. The core our is multi-branch auto-encoder, termed GradNet, which end-to-end learns...

Realistic Rendering of thin transparent layers bounded by rough surfaces involves substantial expense computation time to account for multiple internal reflections. Resorting Monte Carlo rendering such material is usually impractical since recursive importance sampling inevitable. To reduce the burden simulating subsurface scattering and hence improve performance, we adapt microfacet model with a single layer introducing extended normal distribution function (ENDF), new representation this...