Learning Data Generation for Video Digests

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The generation of video digests through deep learning requires a large amount of manual labeling of important scenes, which is labor-intensive and inefficient.

Innovation Solution

A learning data generation device that automatically determines matched sections in raw material data and edited data by verifying feature quantities, generating label data to identify important and non-important sections, thereby reducing the need for manual annotation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual annotation is used to generate learning data, then the quality and accuracy of learning data is improved, but the labor time and cost increase significantly

Engineering Contradiction:

Improveannotation accuracyVSAvoidmanual labeling time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs self-annotation by automatically comparing edited data with raw material data to generate label data, eliminating the need for manual annotation while maintaining high accuracy through feature quantity verification

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system pre-processes the raw material data and edited data by extracting feature quantities before annotation, preparing the data in advance to enable automatic comparison and label generation without manual intervention

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If manual annotation is used to generate learning data, then the quality of learning data is improved, but the productivity decreases due to huge labor requirements

Engineering Contradiction:

Improvelearning data qualityVSAvoiddata generation efficiency

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The annotation process is automated through self-service mechanisms where the system compares feature quantities of edited and raw material data to automatically generate accurate label data, dramatically improving productivity without sacrificing quality

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The manual mechanical annotation process is replaced with an automated computational system that uses feature quantity verification and data comparison algorithms to generate label data, eliminating human labor while maintaining or improving quality

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Extent of automation

If feature quantity verification is performed to automatically generate label data, then the labor required is reduced, but the device complexity increases

Engineering Contradiction:

Improveannotation automationVSAvoidprocessing system complexity

Core Design Contradiction:

Extent of automationVSDevice complexity

Solution Approach 1:

The complex annotation task is segmented into distinct processing steps: feature quantity extraction, data comparison, matched section determination, and label data generation. This modular approach manages complexity while achieving full automation

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS12169960B2Learning data generation device, learning device, identification device, generation method and storage medium

Publication Date: 2024.12.17 NEC CORP

AI summary

The verification unit 53A is configured to determine a matched section which is included in raw material data Dr and edited data De in common by performing verification of raw material feature quantity Fr that is feature quantity of the raw material data and editing feature quantity Fe that is feature quantity of the edited data De, the raw material data including at least one of video data or audio data. The labeling unit 54A is configured to generate, as label data DL corresponding to the raw material data Dr, information which defines the matched section as an important section and defines a section other than the matched section as a non-important section.