摘要
三维重建是计算机图形学中一个经典问题,其中基于单视角图像的三维重建尤为困难。随着深度学习的发展,深度学习在三维重建领域也得到了及其重要的发展。我们在研究过程中,发现了现有三维重建的网络模型重建结果往往缺乏细节信息。而这些信息往往在输入图像中有源可溯。因此我们尝试将注意力机制引入到三维重建中。最后的结果也说明了注意力机制的有效性。
3D reconstruction is a classic problem in computer graphics,where 3D reconstruction based on single-view images is particularly difficult.With the development of deep learning,deep learning has also gained important development in the field of 3D reconstruction.During the research process,we found that the reconstruction results of the existing 3D reconstruction network models often lack detailed information.And this information is often traceable in the input image.So we try to introduce attention mechanisms into 3D reconstruction.The final result also illustrates the effectiveness of the attention mechanism.
作者
胡飞
叶龙
钟微
张勤
HU Fei;YE Long;ZHONG Wei;ZHANG Qin(Key Lab of Media Audio&Video of Ministry of Education,Communication University of China,Beijing 100024)
出处
《中国传媒大学学报(自然科学版)》
2019年第4期24-30,共7页
Journal of Communication University of China:Science and Technology