Split Prediction for Video Coding Block Partitioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding techniques are inefficient in partitioning coding blocks, leading to increased coding time due to the need to evaluate numerous partition possibilities, especially in versatile video coding (VVC) standards like HEVC and VVC, where greedy search algorithms are time-consuming and not adaptive to video content.
Innovation Solution
The use of a convolutional neural network (CNN) to derive probabilities of split types for partitioning coding blocks, allowing the encoder to skip split types below a predetermined threshold and focus on those above the threshold, thereby optimizing the partitioning process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If greedy search algorithms are used to evaluate all possible partition possibilities, then partitioning accuracy is improved, but coding time increases significantly
Solution Approach 1:
The CNN model performs preliminary analysis of the video content before partitioning decisions are made. By pre-processing the video data and extracting relevant features, the system prepares probability information in advance that guides the partitioning process, avoiding the need to evaluate all possible partitions while maintaining accuracy.
Solution Approach 2:
The patent introduces a CNN-based probability estimation model as an intermediary between the video content and the partitioning decision. This intermediary provides predicted probabilities for different partition types, allowing the encoder to make informed decisions without exhaustive search, thus resolving the contradiction between accuracy and time consumption.
2Measurement precision
If all split types are evaluated during partitioning, then optimal partitioning is achieved, but device complexity increases
Solution Approach 1:
The system performs preliminary probability estimation using CNN before the actual partitioning evaluation. This pre-computation of probabilities allows the encoder to focus only on promising partition types, reducing the number of evaluations needed while maintaining optimality.
Solution Approach 2:
The patent changes the approach from evaluating all partition possibilities to evaluating only those with predicted probabilities above a threshold. By transforming the decision criterion from exhaustive evaluation to probability-based selection, the system reduces complexity while preserving optimality for high-probability cases.
Data Source
AI summary
A method implemented by a coding apparatus. The method includes obtaining probabilities of split types being implemented when partitioning a picture, and skipping one or more of the split types based on the probabilities obtained when a coding block is partitioned during a conversion between a video media file and a bitstream. A corresponding apparatus and non-transitory computer readable medium are also provided.