Decoder-Side Neural Network Selection for Video Bitstreams
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current multimedia transport systems lack an efficient method for task-dependent selection of decoder-side neural networks, which hinders optimal performance in varying task categories and new tasks not accounted for during design phases.
Innovation Solution
An apparatus and method that organize and select decoder-side neural networks based on task categories, using shared and non-shared parameters, and perform rate-distortion trade-offs to select the optimal neural network for decoding bitstreams, even for new tasks, by associating neural networks with tasks and evaluating performance using distance metrics and ground truth approximations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple decoder-side neural networks are maintained for different tasks, then task-specific performance is improved, but device complexity increases
Solution Approach 1:
The patent segments the neural network parameters into task-specific and task-agnostic components. Each decoder-side neural network is divided into shared parameters (common across all tasks) and task-specific parameters (unique to each task category). This segmentation allows the system to maintain multiple specialized networks while reducing overall complexity through parameter sharing.
Solution Approach 2:
The patent implements universality through shared parameters that are common across multiple task-specific decoder-side neural networks. These shared parameters enable a single network architecture to perform multiple decoding tasks effectively, reducing the need for completely separate networks for each task while maintaining task-specific optimization through task-specific parameters.
2Measurement precision
If task-specific decoder-side neural networks are used, then decoding accuracy for known tasks is improved, but adaptability to new tasks deteriorates
Solution Approach 1:
The patent enables adaptability to new tasks by allowing dynamic modification of task-specific parameters while keeping the shared parameter architecture intact. When a new task category emerges, the system can introduce new task-specific parameters for the new task while reusing the proven shared parameters, thus maintaining decoding accuracy for known tasks while adapting to new tasks without requiring complete retraining.
3Adaptability or versatility
If all decoder-side neural networks are executed, then comprehensive task coverage is improved, but computation efficiency deteriorates
Solution Approach 1:
The patent implements dynamic selection of decoder-side neural networks based on the detected task category. Instead of executing all available neural networks, the system dynamically identifies the current task category and activates only the corresponding task-specific network with its optimized parameters, significantly improving computation efficiency while maintaining comprehensive task coverage through the organized library of task-specific networks.
Data Source
AI summary
Various embodiments provide an apparatus, a method, and a computer program product. The apparatus includes at least one processor; and at least one non-transitory memory comprising computer program code; wherein the at least one memory and the computer program code are configured to, with the at least one processor, cause the apparatus at least to perform: organize plurality of decoders side neural networks based on one or more task categories or one or more tasks; and select a decoder side neural network based at least on the one or more task categories or the one or more task.