分组卷积编码的单视图三维重建算法研究

Research on Single-View 3D Reconstruction Algorithm for Grouped Convolutional Coding

为了进一步提升单视图图像三维重建的精度,通过对当前的先进算法进行研究,提出了一种改进的单视图三维重建网络.该网络的编码器通过改进特征提取网络,获取更加丰富完整、深层次的二维特征.在精炼器的网络架构中引入注意力机制,进一步细化三维特征,使其生成更加精细的三维体素模型.另外在网络中添加阈值调整模块,来弥补不同种类图像之间的差异,以达到更好的重建效果.实验结果表明,该网络在公共数据集ShapeNet上三维重建的整体IoU值达到0.675,在单图像重建方面取得了更好的效果.

In order to further improve the accuracy of 3D reconstruction of single view image,an improved single view 3D reconstruction network was proposed by studying the current advanced algo-rithms.By improving the feature extraction network,the encoder of the network could obtain more abundant,complete and deep two-dimensional features.The attention mechanism was introduced into the network architecture of the refiner to further refine the 3D features and generate a more refined 3D voxel model.In addition,threshold adjustment module was added in the network to make up for the differences between different kinds of images to achieve better reconstruction effect.The experimental.results show that the overall IoU value of 3D reconstruction on the public data set ShapeNet reaches 0.675,and the network achieves better results in single image reconstruction.