Auto Segmentation Using Bounding Box for Food Identification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current food identification technologies, particularly those using deep learning, face challenges in accuracy due to the need for large amounts of manually labeled data, which is time-consuming, costly, and prone to bias. Additionally, segmentation processes require extensive labeling efforts, increasing the time and cost further.

Innovation Solution

The method and apparatus for auto segmentation using a bounding box employ weakly-supervised learning to segment objects labeled with a bounding box. This approach learns a segmentation model by classifying objects and backgrounds within the bounding box of an object image, allowing for the segmentation of objects in new images using the learned model.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual labeling is performed to obtain data for deep learning model training, then data quality and model accuracy are improved, but time consumption and cost increase significantly

Engineering Contradiction:
Improvemodel accuracyVSAvoidlabeling time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary automated segmentation to generate initial labels before any human inspection. This preliminary action creates a draft labeling that captures the majority of cases, allowing human annotators to focus only on correcting errors rather than creating labels from scratch, thereby reducing overall labeling time while maintaining quality

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system uses automatically generated segmentation results to serve as initial labels for itself. The automated segmentation model produces preliminary labels that are then refined through human inspection only when necessary, creating a self-service labeling workflow that minimizes human intervention while preserving accuracy

Inventive Principle:
Principle #25Self-service

2Manufacturing precision

If segmentation is performed by drawing polygons to achieve precise object boundaries, then segmentation accuracy is improved, but the number of clicks and labeling time increase overwhelmingly compared to bounding boxes

Engineering Contradiction:
Improvesegmentation precisionVSAvoidlabeling efficiency
Core Design Contradiction:
Manufacturing precisionVSProductivity

Solution Approach 1:

The system applies partial action by using bounding boxes only for initial coarse segmentation, then adding pixel-level precision only where necessary through selective refinement. This approach achieves sufficient segmentation accuracy for most cases without requiring the full overhead of polygon drawing for every object, balancing precision and efficiency

Inventive Principle:
Principle #16Partial or excessive action

3Reliability

If a large amount of data is collected for deep learning training to improve model performance, then model robustness is improved, but data quality becomes less critical compared to having diverse edge-case data

Engineering Contradiction:
Improvemodel robustnessVSAvoiddata quantity
Core Design Contradiction:
ReliabilityVSQuantity of substance

Solution Approach 1:

The system changes the parameter focus from data quantity to data quality and diversity. Instead of continuously collecting more data, the system prioritizes collecting edge-case data and diverse examples that challenge the model, recognizing that this qualitative improvement in data composition is more effective for model robustness than sheer volume

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4557232A1Methods and apparatuses for auto segmentation using bounding box
Publication Date: 2025.05.21 NUVI LABS CO LTD
  • EP4557232A1 patent drawingFigure 1
  • EP4557232A1 patent drawingFigure 2
  • EP4557232A1 patent drawingFigure 3

AI summary

Provided are a method and an apparatus for auto segmentation using a bounding box. A method for auto segmentation using a bounding box according to one embodiment of the present disclosure comprises receiving a first object image including an object labeled with a bounding box, which is a pre-learning target, learning a segmentation model by classifying an object and a background from the bounding box of the received first object image, and segmenting an object from a second object image, which is an identification target, using the learned segmentation model.