Generation point cloud from single image is a classical problem in computer vision. The learning methods for this task often adopt local distance metrics as loss function, which means the generated points are not easy to meet the overall shape distribution of the target object. To solve this problem, we introduce a voxel reconstruction network with distribution fitting as auxiliary task and propose a novel framework named Voxel-Assisted Points Generation Network(VAPGN). The auxiliary learning with voxel generation makes it easier to capture the shape distribution of objects in the image during the encoder phase, thereby effectively improving the result of point cloud reconstruction. To meet the needs of mobile and embedded applications, a mobile version of the model is also proposed. In the experiments, we verify the feasibility of our network on the ShapeNet dataset. The proposed framework has achieved outstanding performance on the point cloud generation task, comparing with various state-of-the-art methods. INDEX TERMS Point cloud generation, auxiliary learning.
According to the technical requirements of intelligent development of auxiliary combat system, we construct a visual intelligent test platform. A near-real military scene dataset based on physical rendering is built, which contains 11,000 remote sensing images collected by an analog camera taking pictures in different illumination, weather environment, camera shooting angle, and scene scale condition. Besides, we add a natural style transfer module for a single unmodeled military scene image’s multienvironment generation. We conduct experiments to evaluate the stability of several UAV remote sensing image object detection algorithms. Based on the quality and speed value of the tested algorithms, the adaptability scores in different environments are calculated. Furthermore, we propose a comprehensive evaluation index system of military remote object detection based on a hierarchical model. We envision that our comprehensive benchmark will play a role in the evaluation of algorithm capability for military object detection tasks and the improvement of training algorithm capability.
Acquisition of densely-sampled light fields (LFs) is challenging. In this paper, we develop a coarse-to-fine light field angular superresolution that reconstructs densely-sampled LFs from sparsely-sampled ones. Unlike most of other methods, which are limited by the regularity of sampling patterns, our method can flexibly deal with different scale factors with one model. Specifically, a coarse restoration on epipolar plane images (EPIs) with arbitrary angular resolution is performed and then a refinement with 3D convolutional neural networks (CNNs) on stacked EPIs. The subaperture images in LFs are synthesized first horizontally, then vertically, forming a pseudo 4DCNN. In addition, our method can handle large baseline light field without using geometry information, which means it is not constrained by Lambertian assumption. Experimental results over various light field datasets including large baseline LFs demonstrate the significant superiority of our method when compared with state-of-the-art ones.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.