| Title |
Single-view 3D Reconstruction through Self and Cross Attention Mechanisms |
| Authors |
김혜림(Hye Rim Kim) ; 김선빈(Seon Bin Kim) ; 고병철(Byoung Chul Ko) |
| DOI |
https://doi.org/10.5573/ieie.2024.61.11.112 |
| Keywords |
Single-view 3D reconstruction; 3D object reconstruction; Attention mechanism; Implicit field |
| Abstract |
Single-view 3D reconstruction, the task of reconstructing a 3D object from a single 2D image, is an active area of research in computer vision and graphics with broad applications in augmented reality, game development, and other domains. This paper proposes a novel approach to enhance 3D reconstruction performance by incorporating attention mechanisms into a single-view 3D reconstruction framework. The proposed method applies self-attention and cross-attention mechanisms to a baseline 3D generative model consisting of an encoder-decoder architecture. These mechanisms emphasize salient local features within the image while also utilizing global contextual information to improve reconstruction accuracy. Experimental evaluation on the ShapeNet dataset, a widely used benchmark for 3D reconstruction, demonstrates that the proposed method outperforms state-of-the-art approaches on various metrics assessing reconstruction accuracy and fidelity, and also exhibits superior capability in capturing fine-grained details of complex objects. |