ANN Object Recognition Training with Validation Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for training artificial neural networks (ANNs) for object recognition in images are time-consuming, energy-intensive, and often result in unsatisfactory performance due to the need for large amounts of training data, inefficient tagging and labeling, and lack of diverse training images, especially for uncommon objects.
Innovation Solution
A method that includes training an ANN with a database of images, measuring performance using a validation set, detecting mismatches, correlating these to specific characteristics of the training images, and modifying the database to improve performance by adding or selecting images that address these characteristics, using both real and synthetic images generated through 3D simulation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If a large amount of training images is collected and tagged/labelled, then the training data quantity is sufficient, but the time consumption and labor required become huge
Solution Approach 1:
The patent uses synthetic images generated by 3D simulation as copies of real objects to replace the need for collecting and manually labeling large quantities of real images. These synthetic images contain embedded ground truth information automatically, eliminating the time-consuming manual labeling process while providing sufficient training data quantity.
Solution Approach 2:
The patent performs preliminary generation of synthetic training images with pre-embedded labels and annotations through 3D simulation before the actual training process. This preliminary action creates a ready-to-use training dataset that eliminates the need for subsequent manual labeling and preparation work.
2Reliability
If traditional training methods are used without performance analysis, then the training process is simple, but the performance and precision of the trained ANN remain unsatisfactory
Solution Approach 1:
The patent implements a feedback mechanism where the performance of the trained ANN is evaluated on validation images, and mismatch detection identifies specific characteristics of training images that impact performance. This feedback loop enables targeted modification of the training database, improving ANN reliability while reducing unnecessary training iterations and time consumption.
3Reliability
If more training iterations are performed to improve performance, then the ANN performance may improve, but the training process becomes more time-consuming and energy-intensive
Solution Approach 1:
The performance evaluation and mismatch detection mechanism provides feedback that identifies whether additional training iterations are actually needed. By correlating mismatches with specific training image characteristics, the system can make targeted modifications to the training database, avoiding unnecessary training iterations and reducing energy consumption while maintaining ANN performance improvements.
Data Source
AI summary
The invention relates to a method for training an artificial neural network (ANN) for object detection in images. The method includes a training phase of training the ANN with a training database (TDB) that includes training images, and measuring a performance of the trained ANN with a validation database that includes validation images. If the measured performance is not satisfactory, the method further includes an enhancing phase of detecting a mismatch between the validation images and a response of the trained ANN in response to the validation images, correlating the mismatch to a characteristic of the training images that impacts the training of the ANN, modifying the training database according to the characteristic, and training the ANN with the modified training database. The invention further relates to a non-transitory computer program and a device configured to carry out the method, and to an artificial neural network trained with the method.