Neural Network Video Decoding With Stream-Based Model Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for decoding compressed video content using artificial neural networks lack flexibility and efficiency in selecting the appropriate neural network for decoding, leading to suboptimal performance and compatibility issues between encoding and decoding devices.
Innovation Solution
A method and device that utilize an indicator or identifier within the data stream to determine whether the neural network for decoding is part of a predetermined set or encoded within the stream, allowing for dynamic selection and configuration of the appropriate neural network for decoding, including options for local or remote parameter retrieval.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the neural network parameters are encoded in the data stream, then the decoder can independently decode without external resources, but the data stream size increases and transmission bandwidth is consumed
Solution Approach 1:
The patent extracts neural network parameters from the data stream and stores them externally in a database. The data stream only contains a reference identifier to the external database, significantly reducing data stream size while maintaining decoding independence through external parameter retrieval.
Solution Approach 2:
The patent creates a universal parameter database that can serve multiple data streams and decoding operations. Instead of embedding parameters in each data stream, a single external database provides parameters for multiple decoding tasks, reducing redundancy and improving efficiency.
2Quantity of substance
If the neural network parameters are stored in an external database, then the data stream size is reduced, but the decoder requires external resource access which may not be available
Solution Approach 1:
The patent performs preliminary actions by embedding a reference identifier to the external parameter database within the data stream. This allows the decoder to know in advance where to find the parameters, enabling efficient external resource access while maintaining compatibility across different decoding environments.
Solution Approach 2:
The patent introduces an intermediary reference identifier that connects the data stream to the external parameter database. This intermediary element enables the decoder to access external resources without requiring direct integration, improving adaptability while reducing data stream size.
3Adaptability or versatility
If multiple neural networks are maintained for different decoding scenarios, then the system can adapt to various requirements, but the device complexity and memory requirements increase
Solution Approach 1:
The patent uses copying by storing multiple neural network parameter sets in an external database rather than duplicating them in the decoder device. The decoder can access different parameter sets (copies) as needed without maintaining multiple complete network implementations, reducing device complexity while maintaining adaptability.
Solution Approach 2:
The patent moves neural network parameters from the spatial dimension (local device memory) to an external dimension (remote database). This dimensional shift allows the system to maintain multiple parameter sets without increasing local device complexity, as parameters are accessed remotely rather than stored locally.
Data Source
Figure 1~2
Figure 3~7
Figure 8
AI summary
A data stream comprises an indicator (IND) and data (Fnn) representative of a video or audio content. A method for decoding this data stream comprises the following steps: - decoding (E60) the indicator (IND) to determine whether an artificial neural network to be used to decode the representative data (Fnn) is encoded into the data stream or forms part of a predetermined set of artificial neural networks; - decoding (E70) the representative data (Fnn) by means of the artificial neural network. If it is determined, by decoding the indicator (IND), that the artificial neural network forms part of the aforementioned predetermined set, the method comprises decoding (E62) an identifier (Inn) of the neural network. Associated decoding devices and associated data streams are also described.