Attribute-Based Heatmap Cropping for Principal Image Regions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image cropping technologies fail to adaptively determine the principal portion of an image based on its intended purpose, as the definition of a 'principal portion' varies significantly depending on how the image is used.
Innovation Solution
An information-processing device utilizing a machine learning model group to generate intermediate heatmaps based on attributes of an input image, followed by a generation module to combine these heatmaps and a cutout module to extract the principal portion accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single machine learning model is used to generate heatmaps, then the processing speed is fast, but the accuracy of identifying the principal portion varies depending on image attributes
Solution Approach 1:
The system segments the machine learning model functionality into multiple specialized models, where each model is trained to identify specific attributes (e.g., human images, clothes, accessories) in images. The selection unit then selects the appropriate model based on the image attributes, allowing each model to specialize in specific tasks and thereby improving accuracy without requiring a single overly complex model to handle all scenarios.
2Measurement precision
If multiple machine learning models are used to handle different image attributes, then the accuracy of image cropping improves, but the processing time increases
Solution Approach 1:
The selection unit performs preliminary action by determining the image attributes and selecting the appropriate machine learning model before the actual heatmap generation process begins. This pre-selection based on image attributes ensures that the most suitable specialized model is chosen in advance, avoiding the time penalty of running multiple models or the accuracy loss of using a generic model, thereby optimizing both speed and accuracy.
3Adaptability or versatility
If a fixed machine learning model is used for all image types, then the system is simple to operate, but it cannot adaptively determine the principal portion based on different image attributes
Solution Approach 1:
The system implements self-service through the selection unit, which automatically determines image attributes and selects the appropriate machine learning model without requiring user intervention or manual configuration. The system autonomously adapts to different image types by selecting the most suitable specialized model based on detected attributes, thereby achieving high adaptability while maintaining ease of operation, as users simply need to input images without worrying about model selection.
Data Source
AI summary
Provided is an information-processing device including: a CPU; and a memory storing instructions for causing the information-processing device, when executed by the CPU, to: output an intermediate heatmap for input of an input image by using at least one of a plurality of machine learning models; and generate a heatmap based on an attribute of the input image, which is provided independently of the input image, and the intermediate heatmaps.