Object Classification via Two-Stage Determination to Reduce Computational Load
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection methods, such as those using single shot detectors (SSD), require excessive processing for classifying objects into classes, particularly due to non-maximum suppression (NMS) processing of all candidate bounding boxes, which increases computational load and time, especially in resource-constrained environments like in-vehicle networks.
Innovation Solution
An information processing device and method that employs a machine learning model to classify candidate objects by first determining their likelihood of belonging to a class with negative correlation to other classes, reducing the number of objects subjected to further classification processing, thereby minimizing the overall processing load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If non-maximum suppression (NMS) processing is performed on all candidate bounding boxes, then classification accuracy is improved, but processing time and computational load increase significantly
Solution Approach 1:
The patent segments the classification process into two distinct stages: first determining whether candidate objects belong to a first class (or other classes), and then for those not classified in the first stage, determining their classification into remaining classes. This segmentation reduces the number of candidate objects requiring full classification processing, thereby decreasing computational load and processing time while maintaining classification accuracy.
Solution Approach 2:
The patent applies partial action by performing complete classification processing only on a subset of candidate objects (those not classified in the first determination stage), rather than performing exhaustive classification on all candidate bounding boxes. This partial processing approach significantly reduces computational resources required while still achieving accurate classification results.
2Reliability
If classification processing is executed multiple times for all candidate objects, then classification completeness is improved, but the amount of processing and time increase
Solution Approach 1:
The patent divides the classification task into two sequential determination stages. The first determination stage classifies candidate objects into the first class or other classes, and the second determination stage classifies only those candidate objects that were not classified in the first stage. This segmentation ensures classification completeness while avoiding redundant processing of already-classified objects, thereby improving processing efficiency.
Solution Approach 2:
The patent performs the second determination processing only on a partial set of candidate objects (those determined as non-first class in the first stage), rather than repeating full classification processing on all candidate objects. This partial re-processing approach maintains classification completeness while significantly reducing the total amount of processing required.
3Measurement precision
If all candidate bounding boxes are subjected to classification processing, then detection accuracy is improved, but computational load increases
Solution Approach 1:
The patent segments the candidate object set into two groups through the first determination: those classified as first class/other classes and those not classified. Only the second group undergoes the computationally intensive second determination processing. This segmentation maintains detection accuracy for all candidate objects while reducing computational load by avoiding redundant processing of the first group.
Data Source
AI summary
An information processing device includes a processor. The processor obtains an input image, inputs the input image to a machine learning model that executes classification likelihood calculation processing to obtain, for each of candidate objects in the input image, likelihoods of belonging to the plurality of classes, executes first determination on whether or not each of the candidate objects is classified as a first class of the plurality of classes using a likelihood of belonging to the first class that is a likelihood having a negative correlation with likelihoods of belonging to other classes, executes second determination on whether or not each of the candidate objects that have been determined in the first determination as a non-first class is classified as the other classes, and outputting a result of classifying the candidate objects included in the input image using a result of the second determination.