HOI Detection Weighted Factorization for Smaller ML Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional human-object-interaction (HOI) detection systems require large and complex machine learning (ML) models that consume significant time and computing resources for training, leading to inefficiencies and high memory footprints.
Innovation Solution
A HOI detection system utilizing weighted factorization to determine the importance of object, human, and relationship features, reducing the complexity and size of ML models by generating weights for each feature, enabling accurate and efficient HOI detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional large and complex ML models are used for HOI detection, then detection accuracy can be maintained, but training time and computing resource consumption increase significantly
Solution Approach 1:
The patent segments the HOI detection task into multiple independent feature detection components (object feature detector, human feature detector, pose feature detector, relationship feature detector). Each segment processes specific features independently and produces feature scores that are later combined through weighted factorization. This segmentation allows parallel processing and reduces the computational burden on a single large model, thereby reducing training time while maintaining detection accuracy.
2Measurement precision
If traditional large and complex ML models are used for HOI detection, then detection accuracy can be maintained, but model size and memory footprint increase
Solution Approach 1:
The patent divides the monolithic ML model into multiple specialized feature detectors (object feature detector, human feature detector, pose feature detector, relationship feature detector). Each detector is smaller and more specialized than a general-purpose large model. The segmented architecture reduces overall model size and memory footprint while maintaining detection accuracy through the weighted factorization combination of specialized features.
Solution Approach 2:
The patent applies local quality by making each feature detector specialized for its specific task (object detection, human detection, pose estimation, relationship recognition). Each detector is optimized for its local function rather than being part of a general-purpose large model. This specialization allows each component to be smaller and more efficient while the weighted factorization ensures high-quality integrated detection results.
3Measurement precision
If traditional large and complex ML models are used for HOI detection, then comprehensive feature analysis is possible, but computing resource consumption increases
Solution Approach 1:
The patent segments the computational workload across multiple specialized feature detectors that process different feature types independently. This segmentation enables parallel processing and reduces the cumulative computing resource consumption compared to a single large model processing all features sequentially or in a monolithic architecture. The weighted factorization efficiently combines the results with minimal additional computation.
Data Source
AI summary
Implementations include actions of receiving an image, providing a set of features for the image, determining a set of HOIs including one or more HOIs that are potentially represented in the image, providing sets of feature scores by, for each HOI in the set of HOIs, determining, by a first ML model, a set of feature scores for respective features in the set of features, generating, by a second ML model, sets of weights based on the set of HOIs, providing a set of final scores by, for each HOI in the set of HOIs, determining a final score based on a respective set of weights and the set of feature scores, each final score corresponding to a respective HOI in the set of HOIs, and selecting an output HOI for the image from the set of HOIs based on the set of final scores.