We propose a deep learning approach for directly estimating relative atmospheric visibility from outdoor photos without relying on weather images or data that require expensive sensing custom capture. Our data-driven capitalizes large collection of Internet to learn rich scene and varieties. The CNN-RNN coarse-to-fine model, where CNN stands convolutional neural network RNN recurrent network, exploits the joint power support vector machine, which has good ranking representation, features...

Point cloud analysis without pose priors is very challenging in real applications, as the orientations of point clouds are often unknown. In this paper, we propose a brand new point-set learning framework PRIN, namely, Pointwise Rotation-Invariant Network, focusing on rotation-invariant feature extraction analysis. We construct spherical signals by Density Aware Adaptive Sampling to deal with distorted distributions space. addition, Spherical Voxel Convolution and Re-sampling extract...

Detecting 3D objects keypoints is ofgreat interest to the areas of both graphics and computer vision. There have been several 2D keypoint datasets aiming address this problem in a data-driven way. These datasets, however, either lack scalability or bring ambiguity definition keypoints. Therefore, we present KeypointNet: first large-scale diverse dataset that contains 83,231 8,329 models from 16 object categories, by leveraging numerous human annotations. To handle inconsistency between...

Keypoint detection is an essential component for the object registration and alignment. In this work, we reckon keypoint as information compression, force model to distill out important points of object. Based on this, propose UKPGAN, a general self-supervised 3D detector where keypoints are detected so that they could reconstruct original shape. Two modules: GAN-based sparsity control salient distillation modules proposed locate those keypoints. Extensive experiments show our align well...

In this paper, we tackle the problem of category-level 9D pose estimation in wild, given a single RGB-D frame. Using supervised data real-world poses is tedious and erroneous, also fails to generalize unseen scenarios. Besides, requires method be able objects at test time, which challenging. Drawing inspirations from traditional point pair features (PPFs), design novel Category-level PPF (CPPF) voting achieve accurate, robust generalizable wild. To obtain estimation, sample numerous pairs on...

Detecting aligned 3D keypoints is essential under many scenarios such as object tracking, shape retrieval and robotics. However, it generally hard to prepare a high-quality dataset for all types of objects due the ambiguity keypoint itself. Meanwhile, current unsupervised detectors are unable generate with good coverage. In this paper, we propose an detector, Skeleton Merger, which utilizes skeletons reconstruct objects. It based on Autoencoder architecture. The encoder proposes predicts...

Point cloud analysis without pose priors is very challenging in real applications, as the orientations of point clouds are often unknown. In this paper, we propose a brand new point-set learning framework PRIN, namely, Point-wise Rotation Invariant Network, focusing on rotation invariant feature extraction analysis. We construct spherical signals by Density Aware Adaptive Sampling to deal with distorted distributions space. Spherical Voxel Convolution and Re-sampling proposed extract...

Object pose estimation constitutes a critical area within the domain of 3D vision. While contemporary state-of-the-art methods that leverage real-world annotations have demonstrated commendable performance, procurement such real training data incurs substantial costs. This paper focuses on specific setting wherein only CAD models are utilized as priori knowledge, devoid any background or clutter information. We introduce novel method, CPPF++, designed for sim-to-real category-level...

Using the data-mining machine learning technique and non-equilibrium Green's function method in combination with density functional theory, we studied electronic transport properties of organic-inorganic hybrid perovskite MAPbI3. The band structures MAPbI3 from first-principles show that ferroelectric antiferroelectric dipole configurations have very little influence on energy bandgap. Furthermore, investigated tunnel junctions made 48 different metal electrodes, same fixed lattice constant...

3D object detection has attracted much attention thanks to the advances in sensors and deep learning methods for point clouds. Current state-of-the-art like VoteNet regress direct offset towards centers box orientations with an additional Multi-Layer-Perceptron network. Both their orientation predictions are not accurate due fundamental difficulty rotation classification. In work, we disentangle into Local Canonical Coordinates (LCC), scales orientations. Only LCC regressed, while generated...

Point cloud analysis without pose priors is very challenging in real applications, as the orientations of point clouds are often unknown. In this paper, we propose a brand new point-set learning framework PRIN, namely, Pointwise Rotation-Invariant Network, focusing on rotation-invariant feature extraction analysis. We construct spherical signals by Density Aware Adaptive Sampling to deal with distorted distributions space. addition, Spherical Voxel Convolution and Re-sampling extract...

Diagram object detection is the key basis of practical applications such as textbook question answering. Because diagram mainly consists simple lines and color blocks, its visual features are sparser than those natural images. In addition, diagrams usually express diverse knowledge, in which there many low-frequency categories diagrams. These lead to fact that traditional data-driven model not suitable for this work, we propose a gestalt-perception transformer detection, based on an...

Various recent methods attempt to implement rotation-invariant 3D deep learning by replacing the input coordinates of points with relative distances and angles. Due incompleteness these low-level features, they have undertake expense losing global information. In this paper, we propose CRIN, namely Centrifugal Rotation-Invariant Network. CRIN directly takes as transforms local into representations via centrifugal reference frames. Aided frames, each point corresponds a discrete rotation so...