Neural Image Compression via Dynamic Learning Rate Scheduling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional hybrid video codecs are difficult to optimize as a whole, and improvements in one module may not result in a coding gain in overall performance, limiting the effectiveness of existing video compression technologies.
Innovation Solution
The method involves using a neural network for substitutional end-to-end neural image compression (NIC), where an input image is processed to determine a substitute image based on a training model, which is then encoded and mapped to generate a compressed representation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional hybrid video codec optimization is applied, then individual module improvements can be made, but overall coding gain is not achieved due to difficulty in optimizing as a whole
Solution Approach 1:
The patent merges multiple separate modules (encoder, decoder, and substitution modules) into a unified end-to-end neural network framework. This allows the entire compression system to be optimized jointly through single objective function optimization, resolving the contradiction by enabling holistic optimization while maintaining modular architecture benefits.
Solution Approach 2:
The neural network framework is designed to perform multiple functions within a unified structure: substitution operations, encoding, and decoding. This multi-functional design allows simultaneous optimization of all components, achieving overall coding gain while managing complexity through a universal optimization approach.
2Manufacturing precision
If substitutional end-to-end neural image compression is used, then rate-distortion performance is improved, but computational complexity increases due to neural network processing
Solution Approach 1:
The patent changes the operational parameters of the compression system by using learned substitution operations and adaptive neural network processing instead of fixed traditional algorithms. This allows the system to achieve superior rate-distortion performance by dynamically adjusting parameters based on input characteristics, while the neural network efficiently manages the computational complexity.
3Productivity
If end-to-end optimized framework is applied, then compression efficiency is improved, but additional training requirements increase system complexity
Solution Approach 1:
The patent performs preliminary training of the entire end-to-end framework in advance, optimizing all modules simultaneously before deployment. This preliminary action consolidates the training complexity into a one-time setup process, while the resulting model delivers improved compression efficiency during actual operation without requiring additional training.
Data Source
AI summary
Neural network based substitutional end-to-end (E2E) image compression (NIC) being performed by at least one processor and includes receiving an input image to an E2E NIC framework, determining a step size of the input image indicating a learning rate of a training model, determining a substitute image based on the training model, encoding the substitute image in lieu of the input image to generate a bitstream, and mapping the substitute image to the bitstream to generate a compressed representation. Further, step size may be determined by a scheduler and change throughout the training of the training model. The image may also be split into patches for which a scheduler is assigned for each patch and each patch is encoded instead of the entire input image.