Context-Adaptive Video Coding with Delayed Probability Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding techniques face inefficiencies in entropy coding due to latency and performance degradation from infrequent updates of probability models, which affect coding speed and accuracy, particularly in updating contexts after each bin or full block of video data.
Innovation Solution
The proposed solution involves updating probability models after coding a subset of transform coefficients, allowing for more frequent updates without updating after each bin, and using delayed state updates to avoid delays in probability model availability, enabling parallel processing and maintaining coding performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If probability models are updated after each bin, then coding accuracy is improved, but latency increases and processing speed decreases
Solution Approach 1:
The patent segments the probability model updates into different granularities: context models are updated after each bin for high-accuracy contexts, while other probability models are updated after each coefficient or group of coefficients. This segmentation allows the system to maintain coding accuracy where needed while reducing latency and enabling parallel processing in other areas.
2Productivity
If probability models are updated after each full block of video data, then processing speed increases and latency decreases, but coding performance degrades
Solution Approach 1:
The patent applies local quality by differentiating the update frequency of different probability models based on their importance. Context models, which are critical for coding accuracy, are updated after each bin. Other probability models are updated after each coefficient or subset of coefficients. This local differentiation maintains coding performance while enabling parallel processing and improving speed.
3Measurement precision
If probability models are updated after each coefficient, then coding accuracy is maintained, but parallel processing is prevented and processing speed decreases
Solution Approach 1:
The patent introduces dynamic update strategies where the timing of probability model updates is adjusted based on processing needs. Probability models can be updated at different stages: after each bin, after each coefficient, after each subset of coefficients, or after each full block. This dynamic approach allows the system to adapt between accuracy and speed requirements and enables parallel processing when appropriate.
4Measurement precision
If context models are updated after each bin, then coding precision is improved, but device complexity and computational overhead increase
Solution Approach 1:
The patent applies partial action by selectively updating only the necessary probability models at each stage. Instead of updating all probability models after every bin, the system updates context models after each bin while using delayed or batched updates for other probability models. This partial approach maintains coding precision while reducing computational overhead and device complexity.
Data Source
AI summary
In an example, aspects of this disclosure relate to a method of coding data that includes coding a sequence of bins according to a context adaptive entropy coding process. A current coding cycle used to code at least one bin of the sequence of bins includes determining a context for the bin; selecting a probability model based on the context, wherein the probability model is updated based on a value of a previous bin coded with the context and coded at least two coding cycles prior to the current coding cycle; applying the probability model to code the bin; and updating the probability model based on a value of the bin.