Modular Object Detection Model for Adding Classes Without Retraining

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing object detection models face challenges in maintaining detection accuracy when new objects are added or existing objects are removed, often requiring costly retraining and additional training data.

Innovation Solution

An object detection device that utilizes a modular approach, allowing for the addition or removal of detection objects by combining trained modules with an existing object detector without full retraining, using training data with and without position information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If existing object detection models are retrained to add or remove detection objects, then the detection capability is improved, but the training cost and time increase significantly

Engineering Contradiction:

Improvedetection capabilityVSAvoidtraining time

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The object detection model is segmented into multiple independent modules, each responsible for detecting specific objects. This allows individual modules to be added, removed, or modified without retraining the entire model, thereby improving adaptability while reducing training time and computational costs.

Inventive Principle:

Principle #1Segmentation

2Adaptability or versatility

If existing object detection models are retrained to add or remove detection objects, then the detection capability is improved, but the training cost increases

Engineering Contradiction:

Improvedetection capabilityVSAvoidtraining cost

Core Design Contradiction:

Adaptability or versatilityVSLoss of energy

Solution Approach 1:

Inventive Principle:

Principle #1Segmentation

3Adaptability or versatility

If existing object detection models are retrained to add or remove detection objects, then the detection capability is improved, but additional training data is required

Engineering Contradiction:

Improvedetection capabilityVSAvoidtraining data

Core Design Contradiction:

Adaptability or versatilityVSQuantity of substance

Solution Approach 1:

Multiple trained modules are merged or combined to form a comprehensive object detection model. This allows the system to achieve enhanced detection capability by integrating pre-trained modules without requiring additional training data, as each module has already been trained on its specific object class.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS12573083B2Computer-readable recording medium storing object detection program, device, and machine learning model generation method of training object detection model to detect category and position of object

Publication Date: 2026.03.10 FUJITSU LTD

AI summary

A recording medium storing a program for causing a computer to execute processing including: acquiring, from a first model trained based on training data in which the first object is labeled in an image, a first portion specifying a region in an image that includes a first object; generating a third model by combining the first portion and a third portion of a second model being a model that includes a second portion and the third portion and that is trained based on training data in which position information regarding the second object is labeled in an image, the second portion being a portion that specifies a region in an image including a second object, the third portion being a portion that determines a position in an image of a specified region; and outputting a detection result of an object by inputting an image to the third model.