Related Experiment Video
Updated: Jan 18, 2026
05:33
Three-Dimensional Shape Modeling and Analysis of Brain Structures
Published on: November 14, 2019
7.6K
Generating Point Cloud from Measurements and Shapes Based on Convolutional Neural Network: An Application for
Mau Tung Nguyen1,2, Thanh Vu Dang3, Minh Kieu Tran Thi1
1University of Science and Technology, School of Textile-Leather and Fashion, Ho Chi Minh City, Vietnam.
Computational Intelligence and Neuroscience
|October 1, 2019
Summary
This study introduces a novel method for generating 3D point clouds from measurements using a hierarchical neural network. The approach effectively captures 3D shape information, demonstrating high accuracy and visualization quality.
Area of Science:
- Computer Vision
- 3D Shape Analysis
- Machine Learning
Background:
- 3D shape models are typically parameterized using point clouds or meshes.
- Point clouds offer a simpler alternative to meshes for representing 3D shape information.
Purpose of the Study:
- To introduce a new method for generating 3D point clouds from measurements.
- To represent 3D data using a novel 'slice structure' for shape-measurement correspondence.
- To develop a hierarchical learning model compatible with this data representation.
Main Methods:
- A 'slice structure' representation was developed to link 3D data with measurements.
- A hierarchical neural network model was employed for learning.
- Primary slices were generated by matching measurements, followed by Convolutional Neural Network (CNN) tuning of the point cloud.
Main Results:
- The method was tested on a 3D human dataset (1706 examples).
- Achieved an average error of 7.72%.
- Demonstrated effective visualization of generated 3D point clouds.
Conclusions:
- The proposed framework is effective for generating 3D point clouds from measurements.
- Focusing on local features is crucial for accurate 3D shape processing.
- The method shows promise for applications requiring detailed 3D shape reconstruction.
