Three-Dimensional Model Generation from Two-Dimensional Image Sequences Using Machine Learning

Abstract: In this study, a deep learning model was utilized to generate voxel-represented three-dimensional models of some objects using silhouette images of size 128 × 128 captured from four different angles. The proposed model is trained using the ShapeNet dataset. The deep learning model, along with the proposed error function, has been favored to reduce the number of parameters and capture features of different dimensions. A total of 34 691 different data were obtained in seven categories. The performance metrics of the proposed model have been compared with other studies in the literature using the Intersection over Union (IoU) metric. The comparison reveals that the proposed method achieves an IoU score of 0.5283, which outperforms both the 1 image and 5 image input versions of both McRecon and (Perspective Transformer Nets) PTN methods in categories other than the chair category.Cite this article as: M. Dağtekin, B. Aşıroğlu and Ö. C. Turna, "Three-dimensional model generation from two-dimensional image sequences using machinelearning," Electrica, 25, 0014, 2025. doi: 10.5152/electrica.2025.25014.