Synthetic Image Augmentation for Convolutional Neural Network Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The limited size and scope of image sets used for training convolutional neural networks reduce the efficacy of the training process, as generating comprehensive image sets is time-consuming and requires manual acquisition and labeling of a large number of images.
Innovation Solution
A mixed training set is created that includes both synthetic and non-synthetic images, where synthetic images are generated by applying modifications such as reorientation, occlusion, and changes in lighting or perspective to non-synthetic images, allowing for the augmentation of the training set without the need for extensive manual data collection and labeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a comprehensive image set is used for training, then the training efficacy is improved, but the time and resources required for manual acquisition and labeling increase
Solution Approach 1:
The patent creates synthetic copies of images by applying various transformations (geometric transformations, color adjustments, noise addition) to a limited set of real images. These synthetic images serve as proxies for manually collected images, expanding the training dataset without requiring proportional increases in manual data collection and labeling effort
Solution Approach 2:
The patent performs preliminary data preparation by pre-processing a small set of real images through multiple transformations to generate a comprehensive set of synthetic training images before the actual model training begins. This preliminary action creates a robust training dataset that would otherwise require extensive manual collection
2Adaptability or versatility
If the training set size is increased, then the model's ability to handle diverse image modifications is improved, but the complexity of data preparation increases
Solution Approach 1:
The patent implements a universal data augmentation framework that applies multiple types of transformations (geometric, photometric, noise) to a single set of source images, generating a diverse training dataset from a single source. This multi-functional approach creates images with various modifications without requiring separate data collection processes for each modification type
Solution Approach 2:
The patent systematically varies image parameters including geometric transformations (rotation, scaling, flipping), color adjustments (brightness, contrast, saturation), and noise levels to generate synthetic images. By changing these parameters algorithmically, the system creates diverse training examples without manual intervention for each variation
3Quantity of substance
If synthetic images are generated through multiple transformations, then the training data diversity is improved, but the processing time increases
Solution Approach 1:
The patent generates multiple synthetic copies from each source image through various transformations, creating a multiplied training dataset. Each source image produces numerous variants, exponentially increasing training data diversity without requiring proportional increases in processing time for data collection
Data Source
AI summary
In one respect, there is provided a system that may include a processor and a memory. The memory may be configured to store instructions that results in operations when executed by the processor. The operations may include: training a machine learning model by at least processing a training set with the machine learning model, the training set including at least one synthetic image that is generated by applying one or more modifications to a non-synthetic image; determining, based at least on a result of the processing of the mixed training set, that the machine learning model is unable to classify images having a specific modification; and training the machine learning model with additional training data that includes one or more additional synthetic images having the specific modification. Related methods and articles of manufacture are also disclosed.