OR-Gate Residual Connections for Neural Network Scene Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing scene reconstruction methods using convolutional neural networks face challenges in handling errors and voids in image data, particularly due to reflective surfaces and occlusions, leading to inaccurate reconstructions.
Innovation Solution
The integration of OR-like gates in residual connections within convolutional neural networks to correct errors and voids in sensor data, utilizing a synthetic database approach that injects environment-specific noise during training to improve model robustness and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional convolutional neural networks are used for scene reconstruction, then the system can process image data, but errors and voids from reflective surfaces and occlusions cause inaccurate reconstructions
Solution Approach 1:
The patent applies OR-like gates that convert harmful errors and voids in input data into beneficial information for reconstruction. Instead of treating saturated pixels and occlusions as purely negative factors, the OR-gate logic uses these erroneous inputs to preserve or reconstruct missing scene information, effectively converting data degradation into reconstruction advantage.
Solution Approach 2:
The OR-like gates serve as intermediary components between the input sensor data and the reconstruction output. These gates mediate the flow of information by selectively passing or blocking signals based on the presence of valid data, preventing error propagation while preserving useful information from noisy inputs.
2Area of stationary object
If the model processes all input data including noisy regions, then comprehensive scene coverage is achieved, but spurious inferences are generated from erroneous data
Solution Approach 1:
The patent implements local quality control through OR-like gates that treat different spatial regions of the input data differently. Valid regions are processed normally while erroneous regions (saturated pixels, occlusions) are selectively handled by the OR-gate logic, allowing the system to maintain high precision in reconstructed areas without compromising overall scene coverage.
3Loss of information
If traditional AND-gate logic is used in neural network layers, then logical conjunction is performed, but voids in input data propagate through the network causing loss of scene information
Solution Approach 1:
The patent inverts the traditional AND-gate logic by implementing OR-like gates in the neural network architecture. Instead of requiring all inputs to be valid (AND logic), the OR-gates pass information if at least one input contains valid scene data, effectively inverting the logical operation to prevent information loss from propagating through the network.
Data Source
AI summary
Methods, apparatus, systems and articles of manufacture to reconstruct scenes using convolutional neural networks are disclosed. An example apparatus includes a sensor data acquirer to acquire ground truth data representing an environment, an environment detector to identify an environmental characteristic of the environment, a synthetic database builder to apply noise to the ground truth data to form a training set, a model builder to train a machine learning model using the training set and the ground truth data, and a model adjustor to modify the machine learning model to include residual OR-gate connections intermediate respective layers of the machine learning model. The synthetic database builder is further to store the machine learning model in association with the environmental characteristic of the environment.