Neural Network Object Identification Using Modified 2D Image Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural networks are often inaccurate when performing computer vision tasks on higher dimensional images based on lower dimensional images.
Innovation Solution
A system and process that utilizes neural networks to identify objects in 3D images by modifying and inverting features of 2D images, generating 3D images with enhanced accuracy by incorporating additional features from modified 2D images, and iteratively refining these features to improve object identification in 3D scenes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If neural networks perform computer vision tasks directly on lower dimensional images, then computational resources are saved, but accuracy of object identification deteriorates
Solution Approach 1:
The patent applies preliminary action by modifying 2D images with various transformations (rotation, scaling, flipping, cropping) before feeding them to the neural network. This preprocessing step enhances the input data quality, allowing the network to achieve better accuracy without requiring more computational resources for complex 3D processing
Solution Approach 2:
The patent applies dimensionality change by converting 2D images into pseudo-3D representations through modification and inversion operations. This allows the neural network to process images with enhanced dimensional information while still operating efficiently, bridging the gap between 2D computational efficiency and 3D accuracy requirements
2Measurement precision
If neural networks use modified and inverted features from 2D images to generate 3D images, then object identification accuracy improves, but device complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the complex 3D reconstruction task into separate modular operations: image modification (rotation, scaling, flipping, cropping), feature extraction, inversion operations, and object identification. Each module handles a specific aspect of the processing, making the overall system more manageable and maintainable while achieving high accuracy
Solution Approach 2:
The patent introduces intermediary steps between 2D image input and 3D object identification, including modification operations and inversion processes. These intermediaries enhance the information content of the data without requiring a complete complex 3D reconstruction system, thereby improving accuracy while controlling complexity
3Speed
If neural networks perform 3D computer vision tasks without considering camera or data variance, then processing speed improves, but reliability of results deteriorates
Solution Approach 1:
The patent applies parameter changes by systematically varying image parameters (rotation angles, scale factors, flip directions, crop regions) during the modification process. This creates diverse training data that accounts for camera and data variance, improving reliability without significantly impacting processing speed due to the efficiency of these transformations
Data Source
AI summary
Apparatuses, systems, and techniques to identify objects within one or more images. In at least one embodiment, objects are identified in an image using one or more neural networks based, at least in part, on one or more features of the one or more images and one or more features of one or more modified versions of the one or more images.