Graph-Based Video Classification Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video categorization systems face challenges in accurately categorizing videos due to the lack of large, manually labeled training sets and the assumption of category independence, which fails to capture inter-category relationships.
Innovation Solution
A classifier training system that trains unified classifiers by integrating outputs from multiple initial classifiers, utilizing a category graph to account for inter-category relationships and supplementing labeled videos with co-watched and search-based media items to create a larger, more diverse training set.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised learning is used to train classifier functions, then classification accuracy is improved, but the requirement for large manually labeled training sets increases time consumption and reduces productivity
Solution Approach 1:
The system enables automated training set creation by having the computer automatically select, annotate, and organize training videos based on category relationships and initial classifier outputs, eliminating the need for manual human annotation while maintaining high classification accuracy
Solution Approach 2:
The system performs preliminary classification using initial classifiers trained on small labeled datasets before full training, allowing the main training process to focus on refining classifications rather than creating training data from scratch
2Device complexity
If category independence is assumed in classifier training, then device complexity is reduced, but inter-category relationship information is lost reducing measurement precision
Solution Approach 1:
The system segments the classification problem into two stages: first training simple independent initial classifiers, then combining their outputs through a fusion process that incorporates category relationships, thereby managing complexity while improving accuracy
Solution Approach 2:
The system introduces an intermediate fusion process that takes outputs from independent initial classifiers and combines them using category relationship information, serving as a mediator between simple independent classification and complex joint classification
Data Source
AI summary
A classifier training system trains unified classifiers for categorizing videos representing different categories of a category graph. The unified classifiers unify the outputs of a number of separate initial classifiers trained from disparate subsets of a training set of media items. The training process takes into account the relationships that exist between the various categories of the category graph by relating scores associated with related categories, thus enhancing the accuracy of the unified classifiers.