Data Augmentation with Saliency Maps for Image Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Collecting sufficient training data for machine learning models, particularly for images containing noise or analog images, is difficult, leading to lower accuracy in object recognition.
Innovation Solution
Augment training data by superimposing saliency maps on original images to create additional training data, increasing the number of images available for training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If training data is collected manually for machine learning models, then data quality and accuracy can be maintained, but the time and resources required increase significantly
Solution Approach 1:
The system performs preliminary actions by automatically generating saliency maps and augmenting training images before the actual model training begins. This preliminary data preparation work reduces the time needed during the training phase and eliminates the need for manual data collection, while maintaining high data quality through algorithmic saliency detection.
Solution Approach 2:
The system creates copies of existing training images by superimposing saliency maps on them to generate augmented training data. This copying approach allows the model to learn from multiple variations of the same image without requiring manual collection of additional real-world images, significantly reducing data collection time while maintaining training quality.
2Measurement precision
If more training images are collected to improve model accuracy, then recognition performance increases, but the complexity and cost of data collection increase
Solution Approach 1:
Instead of collecting additional real images, the system creates synthetic copies by overlaying saliency maps on existing training images. This approach generates diverse training data without the complexity of manual data collection, annotation, and curation processes, while still providing sufficient variation for accurate model training.
Solution Approach 2:
The system changes parameters of existing images by modifying them with saliency map overlays, creating augmented versions that highlight different features. This parameter transformation approach generates multiple training variations from single source images, reducing the need for extensive data collection while improving model accuracy.
3Reliability
If manual data annotation is performed to ensure high quality training data, then model training accuracy improves, but the time and labor resources required increase
Solution Approach 1:
The system performs self-service by automatically generating saliency maps and creating augmented training images without human intervention. The algorithm independently identifies salient regions and generates enhanced training data, eliminating the need for manual annotation while maintaining high data quality through automated saliency detection and image augmentation processes.
Solution Approach 2:
The system automatically changes image parameters by superimposing saliency maps with adjusted transparency and blending modes, creating varied training examples without manual editing. This automated parameter transformation maintains data quality while dramatically improving productivity by eliminating time-consuming manual annotation processes.
Data Source
AI summary
A computer implemented method trains an image recognition model. A set of processor units creates a saliency map of an original image. The set of processor units superimposes the saliency map on the original image to form an augmented image, wherein the augmented image is used to train the image recognition model.