Hourglass Predictor for Context-Aware 3D Object Generation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing machine learning methods for generating new objects rely solely on similar types of data, neglecting related data that can provide crucial environmental and contextual information, leading to suboptimal results, especially in complex scenarios like dental crown generation.

Innovation Solution

The use of an hourglass predictor, combining an autoencoder and a second encoder, allows for the incorporation of related data formats to generate objects by training on both similar and related data sets, enabling the generation of objects that consider surrounding conditions, such as dental surroundings.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If machine learning methods use only objects of similar type for generation, then the generation process is simple, but important environmental and contextual information is missed

Engineering Contradiction:
Improveenvironmental and contextual informationVSAvoidmodel architecture complexity
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The model is segmented into two distinct encoders: a first encoder that processes objects of similar type (e.g., dental crowns) and a second encoder that processes related objects (e.g., dental surroundings). This segmentation allows each encoder to specialize in processing different types of input data, preserving both object-specific features and environmental context without requiring a single complex encoder to handle all data types

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The hourglass predictor acts as an intermediary structure that receives latent variables from both the first encoder and second encoder, processes them through a shared bottleneck layer, and generates the final output. This intermediary mechanism integrates information from both similar and related objects, ensuring that environmental and contextual information is incorporated into the generation process

Inventive Principle:
Principle #24Intermediary (Mediator)

2Productivity

If manual design methods are used for dental crowns, then unique circumstances can be considered, but the process is time-consuming and dependent on individual judgment

Engineering Contradiction:
Improvedesign speedVSAvoidautomation level
Core Design Contradiction:
ProductivityVSExtent of automation

Solution Approach 1:

The hourglass predictor enables the system to design dental crowns autonomously by automatically processing both crown-specific data and surrounding dental environment data. The model self-adjusts to unique circumstances through its dual-encoder architecture, which captures relevant features from similar objects and related environmental objects, eliminating the need for manual intervention while maintaining adaptability to individual cases

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The model utilizes latent variables as adjustable parameters that encode essential features from training data. By modifying these latent variables based on the specific dental surroundings and crown requirements, the system can generate customized crown designs automatically, achieving both high productivity and adaptability to unique circumstances

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3931803B1Method for generating objects using an hourglass predictor
Publication Date: 2025.12.03 3SHAPE AS
  • EP3931803B1 patent drawingFigure 1
  • EP3931803B1 patent drawingFigure 2
  • EP3931803B1 patent drawingFigure 3

AI summary

Disclosed is a computer-implemented method for generating a 2D or 3D object, comprising training an autoencoder on a first set of training data to identify a first set of latent variables and generate a first set of output data; training an hourglass predictor on a second set of training data, where the hourglass predictor encoder converts a set of related but different training input data to a second set of latent variables, which decode into a second set of output data of the same type as the first set of output data; and using the hourglass predictor to predict a 2D or 3D object of the same type as the first set of output data based on a 2D or 3D object of the same type as the second set of input data.