Reference Picture Derivation for Temporal Motion Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current motion vector prediction techniques in video coding, such as deriving MVP from spatial or temporal neighbors with scaling or zero motion vectors, are inefficient for modeling temporal information in inter predicted blocks.
Innovation Solution
The techniques utilize intersecting motion vectors, temporal motion vector prediction, and reference picture derivation, incorporating both spatial and temporal neighbors, and motion vectors from already decoded pictures to improve MVP derivation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If motion vector predictor is derived from spatial neighbor or temporal neighbor with scaling, then the derivation process is simple, but the temporal information modeling efficiency is poor
Solution Approach 1:
The patent combines spatial neighbor motion vectors and temporal neighbor motion vectors into a unified MVP derivation process. The motion vector predictor is derived by combining information from both spatial neighbors (same picture) and temporal neighbors (other reference pictures), creating a more comprehensive prediction mechanism that leverages multiple data sources simultaneously to improve temporal information modeling efficiency.
Solution Approach 2:
The patent introduces an intermediary reference picture derivation mechanism that uses already decoded reference pictures as intermediaries. The reference picture derivation unit creates intermediate reference pictures from previously decoded pictures, which then serve as mediators for motion vector prediction. This intermediary approach enables more accurate temporal modeling by using processed reference information rather than raw source data.
2Ease of operation
If motion vector predictor is derived from constructive motion vector or zero motion vector, then the derivation process is simple, but the temporal information modeling is inefficient
Solution Approach 1:
The patent implements feedback mechanisms where the motion vector predictor derived from spatial and temporal neighbors is fed back into the prediction process. The reference picture derivation unit uses previously decoded pictures to generate intermediate references that feedback into the current prediction process, continuously improving accuracy. This feedback loop ensures that temporal information is modeled more reliably by leveraging historical prediction results.
Solution Approach 2:
The patent performs preliminary actions by deriving motion vectors from already decoded reference pictures before they are needed for final prediction. The reference picture derivation unit processes and prepares intermediate reference pictures in advance, so that when motion vector prediction is needed, the prepared reference data is immediately available. This preliminary processing improves temporal modeling accuracy by ensuring high-quality reference data is ready beforehand.
3Speed
If traditional MVP derivation techniques are used, then the encoding process is fast, but the bandwidth efficiency is reduced
Solution Approach 1:
The patent changes the parameters used in motion vector prediction by introducing new prediction candidates that combine spatial and temporal neighbor information. Instead of using only traditional spatial or temporal neighbors, the system changes the prediction parameters to include hybrid candidates that leverage multiple reference sources. This parameter change improves bandwidth efficiency by creating more accurate predictions that require fewer bits to encode, while maintaining fast processing speeds through efficient derivation algorithms.
Data Source
AI summary
Example methods, devices, and computer-readable media are described. An example device for decoding video data includes one or more processors configured to determine a current reference picture for a current block of a current picture of the video data based on a reference picture of a collocated block, a reference picture list of the current block, and a picture order count (POC) of a collocated picture. The collocated block is collocated with the current block and is in the collocated picture. The one or more processors are configured to decode the current block based on a motion vector associated with the current reference picture.