Neural Video Decoding with Residual-Adaptive Probability Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing encoding and decoding methods based on neural networks suffer from poor performance and high complexity, particularly in video processing, due to inadequate consideration of residual changes in probability distribution models.
Innovation Solution
The proposed methods involve obtaining a target probability distribution model based on a scale factor to decode and encode video blocks, adjusting the probability distribution parameters to accurately reflect residual changes, thereby improving encoding and decoding performance while reducing bit rate and complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If neural networks are used for encoding and decoding pictures, then encoding and decoding performance is improved, but complexity increases
Solution Approach 1:
The patent segments the probability distribution modeling into multiple stages: initial probability distribution modeling, residual calculation, and refined probability distribution modeling. This segmentation allows the system to use simpler models at each stage while achieving overall high performance, reducing the need for a single complex neural network.
Solution Approach 2:
The patent performs preliminary encoding using an initial probability distribution model before refining the model based on residuals. This preliminary action allows the system to establish a baseline representation and then improve it iteratively, avoiding the need to use a single complex model for the entire encoding process.
2Productivity
If probability distribution models are used for encoding residuals, then encoding efficiency is improved, but accuracy decreases due to inadequate consideration of residual changes
Solution Approach 1:
The patent implements feedback by calculating residuals between the initial representation and the actual picture data, then using these residuals to refine the probability distribution model. This feedback loop ensures that the model adapts to actual residual characteristics, improving accuracy while maintaining encoding efficiency.
Solution Approach 2:
The patent makes the probability distribution model dynamic by allowing it to change based on residual characteristics. The model is initially set with standard parameters, then dynamically adjusted based on the calculated residuals, enabling it to adapt to different content types and maintain high accuracy across various scenarios.
Data Source
AI summary
The present disclosure provides a decoding method and apparatus, a coding method and apparatus, and a device. The method comprises: on the basis of on a first code stream corresponding to a current block, acquiring a target probability distribution model corresponding to a first scaling factor; on the basis of the target probability distribution model corresponding to the first scaling factor, decoding a second code stream corresponding to the current block, to obtain a corrected residual feature corresponding to the current block; and, on the basis of the corrected residual feature, determining a reconstructed image block corresponding to the current block. The present disclosure improves the coding performance and the decoding performance.