Citation

Remmach, Hassnae and Abdul Razak, Siti Fatimah and Ullah, Arif and Yogarayan, Sumendra and Sayeed, Md Shohel and Mrhari, Amine (2025) CNN-Based Multi-Output and Multi-Task Regression for Supershape Reconstruction from 3D Point Clouds. Informatica, 49 (5). ISSN 0350-5596

Text Enter short title in File_Properties_Summary.pdf - Published Version Restricted to Repository staff only Download (797kB)

Abstract

In a number of industries, including computer graphics, robotics, and medical imaging, three-dimensional reconstruction is essential. In this research, a CNN-based Multi-output and Multi-Task Regressor with deep learning capabilities is proposed for three-dimensional object reconstruction from 3D point cloud. Our approach is grounded in the original Point Net architecture, which addresses the difficulties associated with convolution when applied to point clouds. Firstly, this paper is modified using a Multi-Output Regressor to accurately recreate Super forms from 3D point clouds. Using this method, we first extract features from the 3D point cloud using Point Net. After that, a Multi-Output Regressor receives these data and uses them to anticipate the Super shape parameters needed to reconstruct the shape. Taking in the data, the Multi-Output Regressor retrieved characteristics from Point Net and simultaneously predicts several outcomes. Second, a Multi-Task Regressor is used to modify the Point Net. The network gains from the capacity to transfer knowledge from one task to another, improving the model's overall performance. The model would forecast the ten parameters needed to create the shape in the case of rebuilding Super shapes. The test findings were better than expected; they are intriguing in terms of prediction accuracy and cost, and they update the result by 80%, which is a good accomplishment for the study.

Item Type:	Article
Uncontrolled Keywords:	3D point cloud, 3D reconstruction
Subjects:	Q Science > QA Mathematics > QA71-90 Instruments and machines
Divisions:	Faculty of Information Science and Technology (FIST)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	05 Mar 2025 06:57
Last Modified:	05 Mar 2025 06:57
URII:	http://shdl.mmu.edu.my/id/eprint/13560

Downloads

Edit (login required)