Fused Saliency Map for Automatic Image Annotation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual image annotation for creating training models is time-consuming and costly, requiring large datasets and continuous model refinement, especially when dealing with different categories of images.
Innovation Solution
A computer-implemented method that generates a fused saliency map by combining an image saliency map, a behavior saliency map, and a historical saliency map to automatically create bounding boxes around objects in images, eliminating the need for initial manual annotation and reducing the requirement for key point information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation is used to create training models, then annotation accuracy can be ensured, but the time consumption and cost increase significantly
Solution Approach 1:
The system performs self-annotation by automatically generating saliency maps from input images without requiring manual human annotation. The saliency map generation process enables the system to identify and annotate key objects and regions autonomously, eliminating the need for human annotators while maintaining annotation quality.
Solution Approach 2:
The patent replaces the mechanical manual annotation process with an automated computational process. Instead of human operators manually drawing bounding boxes or labeling objects, the system uses algorithmic saliency map generation to automatically identify and annotate relevant image regions, substituting human labor with computational automation.
2Reliability
If large amounts of manually annotated data are collected to build robust training models, then model reliability improves, but the cost and time required for data preparation increases
Solution Approach 1:
The system generates its own training data by creating saliency maps from unannotated images. This self-service approach eliminates the need to externally collect and manually annotate large datasets, as the system autonomously produces the annotated training data it requires through automated saliency map generation.
Solution Approach 2:
The patent transforms the data preparation process by changing the fundamental parameter from manual annotation to automated saliency map generation. This parameter change allows the system to produce training data efficiently without the time-consuming manual annotation process, thereby improving data preparation productivity while maintaining data quality.
3Measurement precision
If continuous model training is performed to refine annotations, then annotation quality improves, but the time and computational resources required increase
Solution Approach 1:
The system performs preliminary saliency map generation directly from input images before any training occurs. This preliminary action provides initial annotation quality without requiring iterative training cycles, as the saliency map generation process itself produces high-quality annotations that can be used immediately for training or deployment.
4Measurement precision
If extensive key point information is required to build training models, then model accuracy improves, but the complexity of data collection and processing increases
Solution Approach 1:
The patent extracts only the essential key information needed for annotation by generating saliency maps that highlight the most relevant regions and objects in images. This extraction approach obtains sufficient model accuracy by focusing on salient features rather than collecting extensive comprehensive key point information, thereby reducing data collection complexity.
Data Source
AI summary
A computer-implemented method for annotating images is disclosed. The computer-implemented method includes generating a saliency map corresponding to an input image, wherein the input image is an image that requires annotation, generating a behavior saliency map, wherein the behavior saliency map is a saliency map formed from an average of a plurality of objects contained within respective bounding boxes of a plurality of sample images, generating a historical saliency map, wherein the historical saliency map is a saliency map formed from an average of a plurality of tagged objects in the plurality of sample images, fusing the saliency map corresponding to the input image, the behavior saliency map, and the historical saliency map to form a fused saliency map, and generating, based on the fused saliency map, a bounding box around an object in the input image.