Depth Estimation from technical drawings and 3D Mesh Reconstruction with Deep Learning

The transformation of 2D images into detailed 3D models is a long-standing challenge in computer vision, with growing importance in industries where digital production drive workflows and schematic illustrations are often the primary design assets. Automating this process can significantly accelerate workflows and enable the scalable creation of high-quality 3D assets, while minimizing errors and manual effort. The research explores two main approaches: depth estimation from technical drawings and end-to-end 3D mesh reconstruction from single images. The first method is based on predicting depth maps from line-based 2D views with lack of details and convert them into 3D meshes, preserving proportions and design intent. The secondary approach utilizes advanced deep learning architectures to directly generate 3D models, including surface geometry and textures, from a single RGB images. This approach enables the efficient creation of visually enriched 3D assets, providing designers with a solid starting point for refinement. By combining practical insights from the fashion design domain with state-of-the-art computer vision techniques, this work offers scalable solutions for automating 3D object generation. The findings have broader implications, extending to industries such as manufacturing, gaming, and virtual reality, where fast and accurate 3D reconstruction is essential.