Grouped Image Object Detection via Channel Stacking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for object detection in large image datasets or video streams are time-consuming and resource-intensive due to the sparsity of objects-of-interest, requiring analysis of numerous images to find a few relevant ones.
Innovation Solution
The approach leverages the 'cocktail party effect' by training convolutional neural networks (CNNs) to recognize semantic information from grouped images through channel-wise concatenation, allowing for faster detection by reducing the number of input images and using weighted-sums to extract features, thereby accelerating search speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If regular detection methods analyze each image individually in large datasets, then detection accuracy is maintained, but search speed is slow and resource consumption is high
Solution Approach 1:
The patent combines multiple images into a single grouped image by stacking them along the channel dimension, creating a composite input that allows the model to process multiple images simultaneously in one forward pass, thereby accelerating search speed while maintaining detection accuracy
Solution Approach 2:
The trained model achieves multi-functionality by being able to detect objects across multiple different images within a single inference operation, making the detection system efficient for searching through large datasets without requiring separate analysis for each image
2Productivity
If multiple images are processed individually through the CNN, then detailed feature extraction is achieved, but computational resources are wasted
Solution Approach 1:
By stacking multiple images along the channel dimension to form a grouped image, the patent enables the CNN to process multiple images in a single computational pass, significantly reducing the total number of forward propagation operations required and thereby lowering energy consumption
Solution Approach 2:
The images are pre-processed and stacked into grouped images before being fed to the CNN, preparing the data in a format that maximizes computational efficiency during the detection phase and reduces redundant processing
3Productivity
If the number of input images is reduced by combining them into grouped images, then search speed is accelerated, but detection precision may be compromised
Solution Approach 1:
The patent transitions from processing images sequentially in the time dimension to processing them simultaneously in the channel dimension by stacking images, creating a new dimensional representation that preserves fine-grained information while enabling parallel processing and maintaining detection precision
Data Source
AI summary
Disclosed herein is an improved method for identifying images containing objects-of-interest from a large set of images. The method comprises mixing two or more of the images to create a grouped image and exposing the grouped image to an object detector trained on grouped images to make an initial determination that the grouped image was formed from at least one image containing an object-of-interest. The images which formed the grouped image are then exposed to regular object detectors to determine a classification of the object-of-interest.