Self-Training Object Detection Model Accuracy Refinement

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing object detection models suffer from low accuracy in real-world applications due to the lack of relevance between training data and actual input data, which often focuses on specific attributes.

Innovation Solution

A self-training method that collects multimedia data, detects objects using a first object detection model, selects data with target attributes based on accuracy conditions, and trains the model using this data to improve detection accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If multiple sample data of objects with different attributes are collected to train the object detection model, then the model can cover various application scenarios, but the pertinence for data of objects with specific attributes becomes weak, resulting in low detection accuracy

Engineering Contradiction:
Improvecoverage of application scenariosVSAvoiddetection accuracy for specific attributes
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent segments the training process into two stages: initial training with diverse data to establish basic detection capability, and self-training with attribute-specific data to refine accuracy for particular scenarios. This segmentation allows the model to first achieve broad coverage and then specialize in specific attribute detection.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system performs self-training by automatically selecting sample data based on attribute information and using it to improve the model's performance on specific attributes. The model serves itself by utilizing its own output (attribute information) to guide the selection of training data, eliminating the need for external manual curation of attribute-specific datasets.

Inventive Principle:
Principle #25Self-service

2Measurement precision

If attribute-specific sample data is collected to improve detection accuracy for specific attributes, then detection accuracy improves, but the complexity of data collection and model training increases

Engineering Contradiction:
Improvedetection accuracy for specific attributesVSAvoiddata collection and training complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system automatically selects attribute-specific sample data by utilizing the attribute information output by the object detection model itself. This self-service mechanism eliminates the need for manual data collection and curation processes, significantly reducing the complexity of obtaining attribute-specific training data while still achieving improved detection accuracy.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent performs preliminary actions by first collecting diverse sample data and establishing the initial object detection model before proceeding to the self-training phase. This preliminary setup enables the subsequent automatic selection of attribute-specific data without requiring complex data collection infrastructure from the outset.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4220479B1Self-training method, system, apparatus, electronic device, and storage medium
Publication Date: 2025.04.09 HANGZHOU HIKVISION DIGITAL TECHNOLOGY CO LTD
  • EP4220479B1 patent drawingFigure 1~2
  • EP4220479B1 patent drawingFigure 3~4
  • EP4220479B1 patent drawingFigure 5~6

AI summary

The present disclosure provides a self-training method, a self-training system, a self-training apparatus, an electronic device and a storage medium, and relates to the technical field of artificial intelligence. The method includes: obtaining multimedia data collected by a data collection device, and obtaining attribute information by detecting objects in the multimedia data through a first object detection model; selecting, based on the attribute information of the objects in the multimedia data, data containing an object with a target attribute from the multimedia data as first sample data, where the target attribute is an attribute of the object in the multimedia data whose accuracy of a detection result obtained when the first object detection model detects the objects in the multimedia data meets a preset self-training condition; and training the first object detection model by using the first sample data to obtain a second object detection model. The disclosure can improve the accuracy of the object detection model for object detection.