Enhanced 3D Model Generation from 2D Object Images: A Deep Learning Approach Integrating Shape and Texture Learning with Real-Time, Occlusion-Aware and Object-Agnostic Methods

A. Bhagyalakshmi, S. Prabagar, Guruprasad Konnurmath, D. B. K. Kamesh, R. Vishalakshi, Victoriya A.

2025

Abstract

The problem of generating 3D models from 2D object images has been very difficult because it is very complex to capture real shape and texture. The work presents an improved deep learning method for 3D model synthesis which addresses the shortcomings of existing approaches. Our approach embeds the shape and texture learning process in a single framework allowing for real time performance and invariance towards 2D occlusion. Unlike prior methods that need category-specific training, our weapon bidder is object-agnostic and works for various objects other than human-centric ones. We also use effective priors and a novel hybrid learning approach to massively reduce computational cost and improve model generalization. The outcome is a fully scalable, real-time system that can handle high-quality decoding of 3D models from a single 2D image, with potential applications to tasks in augmented reality, virtual reality, and product design. Our method also enforces temporal coherence when applied to videos and is device-agnostic, enabling deployment on edge-based inference devices. We demonstrate experimentally on a broad set of object categories and image qualities that the proposed method is effective and scalable.

Paper Citation

in Harvard Style

Bhagyalakshmi A., Prabagar S., Konnurmath G., Kamesh D., Vishalakshi R. and A. V. (2025). Enhanced 3D Model Generation from 2D Object Images: A Deep Learning Approach Integrating Shape and Texture Learning with Real-Time, Occlusion-Aware and Object-Agnostic Methods. In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies - Volume 1: ICRDICCT`25; ISBN 978-989-758-777-1, SciTePress, pages 71-77. DOI: 10.5220/0013857700004919

in Bibtex Style

@conference{icrdicct`2525,
author={A. Bhagyalakshmi and S. Prabagar and Guruprasad Konnurmath and D. Kamesh and R. Vishalakshi and Victoriya A.},
title={Enhanced 3D Model Generation from 2D Object Images: A Deep Learning Approach Integrating Shape and Texture Learning with Real-Time, Occlusion-Aware and Object-Agnostic Methods},
booktitle={Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies - Volume 1: ICRDICCT`25},
year={2025},
pages={71-77},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013857700004919},
isbn={978-989-758-777-1},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies - Volume 1: ICRDICCT`25
TI - Enhanced 3D Model Generation from 2D Object Images: A Deep Learning Approach Integrating Shape and Texture Learning with Real-Time, Occlusion-Aware and Object-Agnostic Methods
SN - 978-989-758-777-1
AU - Bhagyalakshmi A.
AU - Prabagar S.
AU - Konnurmath G.
AU - Kamesh D.
AU - Vishalakshi R.
AU - A. V.
PY - 2025
SP - 71
EP - 77
DO - 10.5220/0013857700004919
PB - SciTePress