Neural Network Object Re-Inference Based on Size Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network-based systems for object inference in images are computationally expensive and inefficient, particularly when dealing with multiple frames of the same object in videos, leading to significant memory, time, and computing resource usage.
Innovation Solution
Implementing a processor with circuits that execute an algorithm to reduce the number of times a tracked object is re-inferenced by applying criteria such as changes in object size, with dynamic adjustments based on computing resource availability and other factors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If neural networks continuously re-inference tracked objects across multiple video frames, then classification accuracy is improved, but computing resources and time consumption increase significantly
Solution Approach 1:
Instead of continuous re-inference, the system performs inference periodically based on detected events. The neural network is triggered to re-inference objects only when specific conditions are met, such as object size changes exceeding a threshold or when objects enter new regions of interest, thereby reducing redundant computations while maintaining classification accuracy.
Solution Approach 2:
The system uses the output of the neural network itself to determine when re-inference is necessary. By monitoring changes in object characteristics (size, position) and using these changes as triggers for re-inference, the system creates a self-regulating mechanism that avoids unnecessary computational operations while ensuring accurate classification when needed.
2Measurement precision
If the system re-inferences all tracked objects in every frame, then classification refinement is improved, but GPU resource usage becomes excessive
Solution Approach 1:
The system applies different inference strategies to different objects based on their individual characteristics and current state. Objects that meet specific criteria (such as significant size changes or entry into monitored regions) receive re-inference treatment, while objects that do not meet the criteria are excluded from re-inference, thereby optimizing GPU resource allocation.
Solution Approach 2:
The system dynamically adjusts the set of objects subject to re-inference based on real-time conditions. The criteria for re-inference are not fixed but adapt to the current video frame analysis, allowing the system to respond to changing scenes and object behaviors while controlling computational resource consumption.
3Reliability
If the system tracks and re-inferences all detected objects, then object detection completeness is improved, but memory consumption and processing time increase
Solution Approach 1:
The system extracts and processes only the most relevant information about tracked objects. By filtering objects based on criteria such as size changes or regional presence, the system maintains detection completeness for important objects while excluding less relevant objects from detailed re-inference processing, thereby reducing memory and computational requirements.
Data Source
AI summary
Apparatuses, systems, and techniques to use data obtained from an inferred object to determine whether to re-infer the same inferred object. In at least one embodiment, one or more objects are identified in one or more images. A size of the one or more objects in one image is compared to a size of the one or more objects in another image. The one or more objects are re-inferenced based, at least in part, on the comparison.