Convolutional Neural Network Mask Pixel Thickness
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Convolutional neural networks face a trade-off between object detection performance and algorithm complexity, particularly in instance segmentation, where distinguishing objects from the background and improving learning efficiency are challenging.
Innovation Solution
The proposed solution involves modifying the pixel thickness of a real mask used for learning in a convolutional neural network system, incorporating a predicted mask and a real mask with additional pixels for the fake boundary, to enhance learning efficiency and object detection performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If instance segmentation is performed to improve object detection performance, then the ability to distinguish objects from background and each other is improved, but the number of operations and learning complexity increase significantly
Solution Approach 1:
The patent changes the parameter of mask pixel thickness from thin (1-pixel) to thick (multiple pixels) to improve learning efficiency. This parameter change allows the network to learn boundary features more effectively without increasing algorithmic complexity, resolving the contradiction between detection performance and computational burden
Solution Approach 2:
The patent applies preliminary smoothing operations to create thick masks before training the network. By pre-processing the mask data to include fake boundaries and smoothing transitions, the network learns from already-optimized inputs, reducing the learning complexity while maintaining high detection accuracy
2Reliability
If more operations and learning are utilized to improve instance segmentation efficiency, then the ability to distinguish objects is improved, but the training time and computational resources increase
Solution Approach 1:
By changing the mask thickness parameter and adding fake boundaries, the network learns more effective features in fewer iterations. The thick masks provide richer spatial information that accelerates convergence, reducing training time while improving segmentation reliability
Solution Approach 2:
The patent creates fake boundaries by copying and extending the real object boundaries outward. These synthetic training examples provide additional learning samples without requiring more real data, improving segmentation efficiency while avoiding the time cost of collecting and processing additional real-world data
Data Source
AI summary
Disclosed is a computer-readable medium including a program code that, when executed by processing circuitry, causes the processing circuitry to generate a feature map from an input image, to extract a region of interest from the feature map, and to generate a predicted mask based on the region of interest. The processing circuitry may use a predicted mask and a real mask to learn a convolutional neural network system. The real mask includes first pixels corresponding to the real boundary and second pixels corresponding to a fake boundary adjacent to the real boundary.