Neural Network Metadata Signaling for Interoperable Media Streaming
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies lack standardized formats for efficiently exchanging and processing neural network representations across different platforms and environments, particularly for compressed neural networks.
Innovation Solution
The development of a metadata box for neural network representation (NNR) item data, defining associations between NNR item data and configuration using a configuration item property, and creating media files that include NNR tracks, media tracks, NNR media, and media handlers to facilitate storage and streaming of NNR data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If standardized formats are provided for exchange of neural network representations, then interoperability is improved, but device complexity increases
Solution Approach 1:
The patent applies universality by creating a standardized media format (ISOBMFF-based) that can universally represent neural network representations across different platforms and environments. The format defines universal structures including metadata boxes, configuration item properties, and media tracks that serve multiple purposes: storing neural network data, signaling configuration information, and enabling interoperability between different systems without requiring platform-specific custom formats.
2Loss of information
If metadata boxes and configuration item properties are defined for NNR item data, then information completeness is improved, but device complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the neural network representation into distinct structural components: metadata boxes for configuration information, NNR item data for the actual neural network parameters, and association mechanisms (configuration item properties) that link these segments. This segmentation allows each component to be independently processed, stored, and signaled, reducing the complexity of handling the overall structure while ensuring complete information preservation through systematic organization.
3Ease of operation
If media files with NNR tracks and media tracks are defined, then ease of operation is improved, but device complexity increases
Solution Approach 1:
The patent applies merging by combining neural network representation storage with existing media file formats (ISOBMFF). The NNR tracks are integrated into the media file structure alongside regular media tracks, allowing neural network data to be stored, sought, and streamed using the same operational mechanisms as conventional media files. This merging leverages the成熟 infrastructure of media players and processing systems, improving ease of operation while managing complexity through reuse of existing frameworks.
Data Source
AI summary
A method is provided for defining a metadata box of a neural network representation (NNR) item data, wherein the NNR item data comprises an NNR bitstream; and defining an association between the NNR item data and an NNR configuration by using a configuration item property, wherein the NNR configuration item property comprises information about stored NNR item data. Corresponding apparatuses and computer program products are also provided.