Unified Neural Network Model for Multi-Vertical Object Recognition

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing object recognition systems require multiple specialized models for different object verticals, leading to substantial resource consumption and computational demands, especially on mobile platforms.

Innovation Solution

A unified machine learning model is generated using a deep neural network that combines learning targets from specialized models trained with a triplet loss function, employing an L2-loss function for training to reduce complexity and improve efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If multiple specialized models are trained for different object verticals, then object recognition accuracy for each vertical is improved, but model storage resources and computational demands increase substantially

Engineering Contradiction:
Improveobject recognition accuracyVSAvoidmodel storage resources and computational demands
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent merges multiple specialized models into a single unified model that can recognize objects across different verticals (e.g., apparel, footwear, accessories). The unified model integrates the learning targets from multiple specialized models trained with triplet loss function, enabling it to perform object recognition tasks that previously required separate specialized models while reducing overall system complexity and resource consumption.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The unified model is designed with universal capability to handle multiple object verticals simultaneously. By incorporating learning targets from various specialized models, the single unified model can perform object recognition, retrieval, and identification functions across different categories (apparel, footwear, accessories) without requiring separate specialized models for each vertical, thus achieving multi-functionality.

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Reliability

If separate specialized models are deployed for different object verticals, then recognition performance for each vertical is optimized, but processor utilization and computational footprint increase

Engineering Contradiction:
Improverecognition performanceVSAvoidprocessor utilization and computational footprint
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

The patent combines multiple specialized models into a unified model structure that maintains recognition performance across different verticals while reducing computational footprint. The unified model processes images through a single neural network architecture that integrates learning targets from multiple specialized models, eliminating the need to run multiple separate models and thereby reducing processor utilization and energy consumption.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent employs parameter changes in the loss function (using L2-loss instead of triplet loss) to enable the unified model to learn from multiple verticals effectively. This parameter change allows the model to optimize its internal representations to capture vertical-specific characteristics while maintaining a unified architecture, thus preserving recognition performance with reduced computational requirements.

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If multiple specialized models are used for object recognition, then accuracy for specific verticals is improved, but system scalability and deployment efficiency decrease

Engineering Contradiction:
Improvevertical-specific recognition accuracyVSAvoidsystem scalability and deployment efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent merges multiple specialized models into a single unified model that scales more efficiently. Instead of deploying and managing multiple separate models for different verticals, the unified model provides a single scalable architecture that can handle various verticals through its integrated learning targets, improving deployment efficiency and system scalability while maintaining vertical-specific recognition accuracy.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP3642764B1Learning unified embedding
Publication Date: 2025.03.12 GOOGLE LLC
  • EP3642764B1 patent drawingFigure 1
  • EP3642764B1 patent drawingFigure 2
  • EP3642764B1 patent drawingFigure 3

AI summary

A computer-implemented method for generating a unified machine learning model using a neural network on a data processing apparatus is described. The method includes the data processing apparatus determining respective learning targets for each of a plurality of object verticals. The data processing apparatus determines the respective learning targets based on two or more embedding outputs of the neural network. The method also includes the data processing apparatus training the neural network to identify data associated with each of the plurality of object verticals. The data processing apparatus trains the neural network using the respective learning targets and based on a first loss function. The data processing apparatus uses the neural network trained to generate a unified machine learning model, where the model is configured to identify particular data items associated with each of the plurality of object verticals.