Entropy Coding for Neural Media Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural network variables related to entropy coding, when quantized, lead to significant degradation in compression efficiency, especially in common use cases, and these losses cannot be recovered by re-training the neural network.
Innovation Solution
Optimizing the definition of trained entropy coding variables to preserve important information when represented with low-precision integers, using a logarithmic function of the standard deviation as a probability distribution function parameter, and determining a code vector based on this parameter for entropy coding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If neural network variables for entropy coding are quantized to low-precision integers, then speed and power consumption are improved, but compression efficiency degrades significantly
Solution Approach 1:
The patent applies parameter changes by transforming the probability distribution function parameter from a linear scale to a logarithmic scale. This transformation optimizes the representation of entropy coding variables when quantized to low-precision integers, preserving compression efficiency while enabling faster, lower-power execution with integer arithmetic
2Speed
If neural network variables for entropy coding are quantized to low-precision integers, then processing speed is improved, but compression efficiency degrades significantly
Solution Approach 1:
The patent transforms the probability distribution function parameter using a logarithmic function, which optimizes the quantized representation and maintains compression efficiency when using low-precision integers for fast processing
3Ease of manufacture
If general quantization optimization tools are used for neural networks, then implementation is simplified, but the specific properties of entropy coding variables are not taken into account
Solution Approach 1:
The patent applies local quality by tailoring the quantization optimization specifically to entropy coding variables. Instead of using general quantization tools, the invention customizes the probability distribution function parameter transformation to account for the specific requirements of entropy coding, thereby minimizing information loss
4Device complexity
If standard deviation is used directly as probability distribution function parameter, then the definition is simple, but quantization effects cause worst compression losses
Solution Approach 1:
The patent transforms the standard deviation parameter through a logarithmic function to create an optimized probability distribution function parameter. This transformation simplifies the quantization process and minimizes compression losses while maintaining ease of implementation
Data Source
AI summary
This disclosure describes entropy coding techniques for media data coded using neural-based techniques. A media coder is configured to determine a probability distribution function parameter for a data element of a data stream coded by a neural-based media compression technique, wherein the probability distribution function parameter is a logarithmic function of a standard deviation of a probability distribution function of the data stream, determine a code vector based on the probability distribution function parameter, and entropy code the data element using the code vector.