3D Model Component Search with Embedding-Based Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Identifying specific components within three-dimensional models is challenging due to the large number of models in data stores and the difficulty in knowing where the components are located or how to search for them effectively using text-based queries.
Innovation Solution
A computing device generates representation vectors from 2-D images of 3-D models and components using a hybrid machine-learning model, indexes these vectors, and compares them to query embeddings to identify similar components, providing a user interface for selection and automatic population of relevant files.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If text-based search queries are used to identify components in 3-D models, then the search process becomes simpler for users, but the ability to accurately identify specific components deteriorates due to the large number of models and components in data stores
Solution Approach 1:
The patent introduces an embedding model as an intermediary that transforms text-based component names into vector representations. This embedding model acts as a mediator between the simple text query interface and the complex 3-D model database, enabling accurate component identification without requiring users to navigate through numerous models directly. The embedding vectors serve as the intermediary representation that bridges the gap between human-readable component names and machine-stored model data.
Solution Approach 2:
The patent transforms the search problem from comparing text strings to comparing vector embeddings in a continuous space. By changing the parameter representation from discrete text to continuous vector embeddings, the system enables more precise similarity matching. The embedding model maps component names to vectors that capture semantic relationships, allowing the search to identify components based on semantic similarity rather than exact text matching, thereby improving identification accuracy while maintaining ease of use.
2Adaptability or versatility
If the data store contains an excessive number of 3-D models and components, then the comprehensiveness of the model library improves, but the difficulty of locating specific components increases
Solution Approach 1:
The patent replaces the mechanical browsing approach (manually navigating through models and components) with an automated semantic search system. Instead of requiring users to mechanically search through the excessive number of models, the system uses embedding-based semantic matching to automatically locate components. This substitution transforms the search mechanism from a manual exploration process to an automated retrieval process based on semantic understanding of component names.
Solution Approach 2:
The embedding model serves as an intermediary layer between the user's text query and the large database of 3-D models. This intermediary transforms the unstructured text query into structured vector representations that can be efficiently compared against the embedded component data, enabling the system to handle the excessive number of models without increasing user difficulty in locating specific components.
3Device complexity
If traditional search methods are used to find components in 3-D models, then the system structure remains simple, but the search efficiency and accuracy deteriorate
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing embedding vectors for all components in the 3-D models during an offline indexing phase. This preliminary processing creates a ready-to-use semantic index that enables fast online search operations. By performing the computationally intensive embedding generation in advance, the system maintains simple interaction structures during user searches while achieving high search efficiency through pre-prepared vector comparisons.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computing device can receive 3-D model data that includes components included in 3-D models. The components can be represented by 2-D images. The computing device can generate representation vectors that correspond to different 2-D images. The computing device can index the representation vectors to facilitate queries with respect to the components. The computing device can receive a query indicating a request to identify 2-D images associated with the query. The computing device can compare an embedding of the query to the representation vectors to identify a set of representation vectors that match the embedding. The computing device can provide, on a digital user interface, the 2-D images corresponding to the set of representation vectors.