Neural Image Decoding With Grouped Convolution for Mobile Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image encoding/decoding solutions in neural network models have high decoding complexity and low efficiency, making them unsuitable for mobile devices.
Innovation Solution
Implement a decoding neural network with decoding indication information to simplify decoding operations, including grouping convolution operations and channel reconstruction, reducing complexity and improving efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a neural network model is used for end-to-end image compression, then image compression can be achieved, but decoding complexity increases and decoding efficiency decreases
Solution Approach 1:
The patent divides the decoding neural network into multiple convolutional layers, each processing specific feature maps. The grouping convolution operation segments the channel dimensions into multiple groups, allowing parallel processing and reducing computational complexity while maintaining decoding efficiency.
Solution Approach 2:
The patent applies different convolution operations to different feature maps based on their specific requirements. The grouping convolution allows each group to be processed with appropriate kernels and parameters, optimizing local processing quality while reducing overall complexity through selective computation.
2Ease of operation
If standard convolution operations are used in the decoding neural network, then complete feature processing can be achieved, but computational resources are over-consuming for mobile devices
Solution Approach 1:
The patent changes the convolution operation parameters by introducing grouping, where channels are divided into multiple groups. This parameter change reduces the number of computations required while maintaining the essential feature extraction capabilities, making the model suitable for mobile devices with limited computational resources.
Solution Approach 2:
The patent applies partial convolution by processing only the necessary portions of feature maps through grouping. Instead of processing all channels uniformly, the grouped convolution selectively processes relevant features, reducing computational overhead while maintaining adequate processing capability for mobile applications.
Data Source
AI summary
Embodiments of this application provide an image processing method performed by a computer device. The method includes: obtaining a bit stream of an image; extracting, from the bit stream, decoding indication information, the decoding indication information indicating a decoding operation to be performed in a decoding neural network, and invoking, based on the decoding indication information, a decoding neural network to perform the decoding operation on the bit stream to reconstruct the image, wherein the decoding operation is configured for performing grouping on a convolution operation in at least one convolutional layer in the decoding neural network. The embodiments of this application can reduce decoding complexity on a decoder side, thereby improving decoding efficiency.