基于改进卷积神经网络的单幅图像物体重建方法

为了提高基于图像的三维重建的重建效果,基于深度学习的方法已经成为近年来研究的重点。针对目前存在的方法中特征提取效果差、重建细节缺失且计算量巨大的问题,提出一种改进卷积神经网络的单个物体重建方法。通过加入改进的Inception resnet模块来提升网络的特征提取能力,采用多种网络结构提取多特征,通过多特征依次输入3D-LSTM模块中以增强单幅图像的重建效果。实验结果表明,该方法不仅能够得到更好的重建效果,重建出更多的细节,同时在训练中花费更少的时间。

SINGLE IMAGE OBJECT RECONSTRUCTION METHOD BASED ON IMPROVED CONVOLUTIONAL NEURAL NETWORK

To improve the performance of 3D reconstruction,methods based on deep learning have been the main topic of the research.Aiming at the problems of poor feature extraction effect,lack of reconstruction detailsand huge computational load,we proposed a 3D objects reconstruction method from a single image based on improved convolutional neural network.The feature extraction capability of the network was improved by adding modules that combine residual connectionsand Inception.We used multi-features extracted by multi-network structure,and input it into 3D-LSTM module in turn to enhance the reconstruction effect of a single image.The experimental results show that our method can not only perform better in reconstruction,but also spend less time in training.