Video Subsegment Selection for ML Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training machine learning models for video analytics using entire video segments is computationally intensive and costly, requiring large storage and processing resources.
Innovation Solution
Selecting subsets of video subsegments based on specific criteria, such as random selection, equal spacing, or modified object characteristics, for training ML models, which reduces storage and processing demands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If entire video segments are used for training ML models, then model accuracy is improved, but computational intensity and storage requirements increase
Solution Approach 1:
The patent divides video segments into multiple frames and selects specific subsets of frames for training instead of using entire video segments. This segmentation approach maintains model accuracy by preserving key information while reducing computational intensity and storage requirements by processing only selected frames.
Solution Approach 2:
The patent applies partial action by selecting a subset of frames rather than processing the complete video segment. The selection criteria ensure that the most informative frames are chosen, providing sufficient training data for accurate model learning without the excessive computational burden of processing all frames.
2Measurement precision
If entire video segments are used for training ML models, then model accuracy is improved, but storage requirements increase
Solution Approach 1:
The patent segments video data into individual frames and stores only the selected subset of frames for training purposes. This reduces storage requirements while maintaining model accuracy by preserving the most informative frames needed for effective model training.
Solution Approach 2:
The patent extracts and stores only the essential frames from video segments based on selection criteria. This extraction process reduces storage requirements by eliminating redundant frames while retaining the key information necessary for accurate model training.
3Use of energy by moving object
If subsets of video frames are selected for training, then computational intensity is reduced, but model accuracy may deteriorate
Solution Approach 1:
The patent applies local quality by using different selection criteria for different video segments based on their characteristics. This ensures that each segment contributes its most valuable frames to the training set, maintaining overall model accuracy while reducing computational intensity through targeted frame selection.
Solution Approach 2:
The patent changes parameters by adjusting frame selection criteria, spacing, and subset sizes to optimize the balance between computational intensity and model accuracy. These parameter adjustments ensure sufficient training data is provided while keeping computational requirements manageable.
Data Source
AI summary
A management node is described. A method implemented in a management node is described. The method comprises receiving a video segment and selecting a plurality of subsegments of the video segment based on a selection criteria. The selection criteria comprise rules based on at least one of a selection characteristic, a subsegment characteristic, a video segment characteristic, or a system characteristic. The method further includes training a machine learning (ML) model using the plurality of subsegments selected based on the selection criteria.