Template-Based Inter Prediction Latency Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding standards, such as HEVC, face challenges in reducing decoding latency due to the dependency on fully reconstructed neighboring blocks for template-based inter prediction methods, which complicates pipeline design and increases hardware implementation complexity.
Innovation Solution
The proposed solution involves using prediction signals from neighboring blocks as template samples for template-based inter prediction, allowing decoding to start before the reconstruction of neighboring blocks is complete, and dividing a picture into template slices to enable independent decoding of inter blocks within a slice.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If template-based inter prediction uses fully reconstructed neighboring blocks as templates, then prediction accuracy is improved, but decoding latency increases and pipeline complexity increases
Solution Approach 1:
The patent applies preliminary action by using prediction signals from neighboring blocks that are available before complete reconstruction is finished. The template is constructed from prediction signals of neighboring blocks at an earlier stage in the decoding pipeline, allowing the current block to be decoded in parallel without waiting for full reconstruction of neighbors, thus reducing latency while maintaining acceptable prediction accuracy
Solution Approach 2:
The patent segments the picture into template slices that are processed independently. Each template slice can be decoded in parallel without waiting for other slices to be fully reconstructed, enabling pipelined processing. This segmentation allows the decoding system to proceed with multiple blocks simultaneously, reducing overall decoding latency while maintaining prediction quality within each slice
2Measurement precision
If template-based inter prediction waits for full reconstruction of neighboring blocks, then prediction quality is improved, but hardware implementation complexity increases
Solution Approach 1:
The patent uses prediction signals that are generated earlier in the decoding process rather than waiting for complete reconstruction. This allows the hardware pipeline to proceed in parallel without requiring complex synchronization mechanisms or buffer management, simplifying the hardware implementation while maintaining prediction quality
Solution Approach 2:
By dividing the picture into independent template slices, the patent enables simple parallel hardware processing where each slice can be handled by dedicated processing units without complex coordination. This segmentation approach simplifies hardware design compared to a monolithic approach that would require complex inter-block synchronization and data sharing mechanisms
3Productivity
If template-based inter prediction is applied to reduce coding overhead, then coding efficiency is improved, but decoding latency increases
Solution Approach 1:
The patent performs template construction using prediction signals at an earlier stage in the decoding process, before complete reconstruction of neighboring blocks is finished. This allows the template-based inter prediction to be executed in parallel with other decoding operations, maintaining coding efficiency while reducing the time required for decoding completion
Solution Approach 2:
The patent segments the decoding process into independent template slices that can be processed in parallel. This segmentation enables the system to maintain high coding efficiency through template-based prediction while reducing decoding latency by processing multiple slices simultaneously without sequential dependencies
Data Source
Figure 1A
Figure 1B
Figure 2
AI summary
Video coding methods are described for reducing latency in template-based inter coding. In some embodiments, a method is provided for coding a video that includes a current picture and at least one reference picture. For at least a current block in the current picture, a respective predicted value is generated (e.g. using motion compensated prediction) for each sample in a template region adjacent to the current block. Once the predicted values are generated for each sample in the template region, a process is invoked to determine a template-based inter prediction parameter by using predicted values in the template region and sample values the reference picture. This process can be invoked without waiting for reconstructed sample values in the template region. Template-based inter prediction of the current block is then performed using the determined template-based inter prediction parameter.