3D Geometric Model Tensor Pre-processing for Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional geometric model data in 3D modeling software is complex and large, leading to slow feature extraction when used in deep learning, and converting it to 2D format results in inaccurate image classification due to loss of details.
Innovation Solution
A method and system for pre-processing geometric model data by creating a 3D geometric model tensor using a processor, which involves assigning properties to virtual grids based on object dimensions and replacing initial values with identification attributes, allowing direct input into deep learning modules.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If conventional geometric model data is used directly for deep learning, then the model contains complete 3D spatial information, but the feature extraction takes a long time due to large data size and complexity
Solution Approach 1:
The patent segments the continuous 3D geometric model data into discrete voxel grids, where each voxel represents a discrete spatial unit. This segmentation transforms the complex continuous spatial data into a structured grid format that deep learning models can process efficiently while preserving the essential 3D spatial relationships and object properties.
Solution Approach 2:
The patent changes the parameter representation by converting traditional geometric model parameters (coordinates, dimensions, orientations) into a voxel-based parameter system. Each voxel is assigned parameters such as material type, object ID, and spatial position, transforming the data structure into a format optimized for deep learning processing while maintaining the underlying geometric information.
2Loss of time
If geometric model data is converted to 2D format for deep learning input, then the data size is reduced, but the classification accuracy decreases due to loss of object details
Solution Approach 1:
Instead of reducing to 2D, the patent maintains and utilizes the 3D dimension by creating a volumetric voxel grid representation. This approach preserves the third spatial dimension, allowing deep learning models to access complete 3D spatial information, object heights, and spatial relationships that would be lost in 2D projections, thereby maintaining high classification accuracy while achieving computational efficiency.
3Loss of information
If conventional geometric model data is used, then complete object information is available, but the file size is large and the data structure is complex
Solution Approach 1:
The patent segments the complex geometric model into a regular voxel grid structure, where the space is divided into uniform discrete units. This segmentation simplifies the data structure by replacing complex geometric descriptions with a regular grid of standardized voxels, making the data more manageable and suitable for deep learning while preserving all essential object information through voxel assignments.
Solution Approach 2:
The patent transforms the parameter representation from detailed geometric descriptions (coordinates, surfaces, volumes) to simplified voxel-based parameters (grid position, material ID, object label). This parameter transformation reduces data structure complexity by using a standardized, discrete parameter system that is inherently more suitable for computational processing while maintaining information completeness through the voxel encoding scheme.
Data Source
AI summary
A method for pre-processing geometric model data of a 3D modeling software for deep learning includes steps of: determining a size of a virtual grid that is visualized as a cube based on a smallest value of dimension among values of dimension of objects; generating an empty tensor that is visualized as a cuboid consisting of the virtual grids; assigning an initial value to each of the virtual grids of the empty tensor; replacing the initial value of each of those of the virtual grids with a pre-determined identification attribute value that corresponds uniquely to the property of the corresponding one of the at least one geometric object, so as to generate a 3D geometric model tensor to be used as an input for deep learning; and saving the 3D geometric model tensor in a database in a predefined format.