Coding Unit Depth Prediction Using Pre-Trained Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing HEVC coding standard's traversal method for determining coding unit (CU) depth leads to a long coding prediction time and significant consumption of calculation resources due to its inefficiencies.
Innovation Solution
A CU depth determining method that uses a pre-trained predictive model to analyze residual coefficients and coding information features of CUs and neighboring CTUs, determining whether depth division is needed, thereby reducing unnecessary calculations and optimizing coding processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the existing traversal method is used to determine CU depth, then the coding performance can be optimized by calculating rate-distortion cost for all modes, but the coding prediction time becomes extremely long and calculation resources are significantly consumed
Solution Approach 1:
The patent applies preliminary action by pre-training a predictive model offline using training samples with classification results. This pre-trained model can then quickly predict whether depth division is needed during actual coding, avoiding the time-consuming traversal process while maintaining coding performance. The model is trained beforehand to recognize patterns that indicate whether further depth division will benefit coding efficiency.
Solution Approach 2:
The patent introduces a predictive model as an intermediary between the input video data and the depth division decision. This model acts as a mediator that processes coding information features and residual coefficients to predict the optimal depth division strategy, replacing the direct traversal approach and significantly reducing calculation time while preserving coding performance.
2Reliability
If the existing traversal method is used to determine CU depth, then all modes with different depths can be evaluated for optimal coding, but a large quantity of calculation resources are consumed
Solution Approach 1:
The patent extracts only the essential features needed for depth division prediction, such as coding information features and residual coefficients, rather than processing all possible modes. By extracting these key features and feeding them to the pre-trained predictive model, the system avoids the exhaustive calculation of rate-distortion costs for all depth modes, significantly reducing computation resources while maintaining decision quality.
Solution Approach 2:
The patent replaces the expensive and time-consuming traversal calculation process with a lightweight predictive model that requires minimal computational resources. The model makes quick predictions about depth division needs without performing exhaustive calculations, effectively substituting a resource-intensive process with a more efficient alternative that consumes far fewer calculation resources.
3Reliability
If depth division is performed for all CUs to ensure optimal coding, then coding quality is maintained, but unnecessary depth division and rate-distortion cost calculations increase processing time
Solution Approach 1:
The patent applies local quality by making depth division decisions specific to each CU based on its local characteristics. The predictive model analyzes coding information features and residual coefficients of each CU to determine whether that particular CU needs depth division, rather than applying a uniform approach to all CUs. This localized decision-making maintains coding quality where needed while avoiding unnecessary processing where it is not beneficial.
Solution Approach 2:
The patent uses partial action by performing depth division only for CUs where it is actually needed, as predicted by the model, rather than applying depth division universally. This selective approach performs the necessary action (depth division) only in partial cases where it improves coding quality, while skipping it in cases where it would be excessive and wasteful, thereby improving overall coding efficiency.
Data Source
AI summary
A coding unit (CU) depth determining method includes determining a residual coefficient of a current optimal mode of a target CU, respectively obtaining coding information features of the target CU corresponding to a specific frame type and coding information features of a neighboring coding tree unit (CTU) corresponding to the specific frame type of a CTU in which the target CU is located, to form prediction feature vector samples, and inputting the prediction feature vector samples to a pre-trained predictive model corresponding to the specific frame type, to obtain a prediction result.