Generative AI Media Compression Using Partially Decompressed Syntax Tokens
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training generative AI models with uncompressed media data is challenging due to the immense volume of information and redundant content, which complicates the training process.
Innovation Solution
Using partially decompressed data as input for generative AI models, where syntax elements are converted into tokens and used for training, reducing data volume and simplifying data organization, enabling efficient training and processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If uncompressed media data is used for training generative AI models, then the information content is complete and accurate, but the data volume is immense and contains significant redundant information
Solution Approach 1:
The patent extracts only the essential syntax elements from compressed media data that are necessary for training generative AI models. Instead of using complete uncompressed data, the system identifies and extracts key syntax elements (such as motion vectors, prediction modes, and other structural parameters) that capture the important information content while eliminating redundant pixel-level data. This extraction process resolves the contradiction by maintaining information completeness at the structural level while dramatically reducing data volume.
2Measurement precision
If uncompressed media data is used for training, then all detail information is available, but the training process becomes complicated and less efficient
Solution Approach 1:
The patent segments the media data into discrete syntax elements that can be independently processed and learned by the generative AI model. Instead of presenting the model with overwhelming continuous pixel data, the system divides the information into structured segments (syntax elements) that represent specific aspects of the media content. This segmentation simplifies the training process by making the data more manageable and the learning objectives clearer, while still preserving the necessary detail information through the structured representation of syntax elements.
3Quantity of substance
If compressed data is used directly as input, then data volume is reduced, but the data format is not optimized for generative AI model processing
Solution Approach 1:
The patent introduces syntax elements as an intermediary representation between compressed data and the generative AI model. The compressed data is first decoded into syntax elements, which serve as a bridge that translates the compressed format into a form that is both compact and suitable for AI processing. These syntax elements act as a mediator that preserves the space-efficient benefits of compression while providing a structured, interpretable format that optimizes the ease of operation for training generative models. The syntax elements maintain data volume reduction while improving format compatibility with AI model requirements.
Data Source
AI summary
A computer system performs operations to compress or decompress media using a generative artificial intelligence (“AI”) model. For compression, a compression tool receives a first version of input media, which has a first resolution, converts the first version to a second version, which has a second resolution lower than the first resolution, compresses the second version, and outputs compressed data for the second version. For decompression, a decompression tool receives input tokens that represent input syntax elements, respectively, of the compressed data for the second version. The decompression tool provides the input tokens to the generative AI model and receives predicted tokens from the generative AI model. The predicted tokens represent output syntax elements, respectively, of compressed data for output media. The decompression tool reconstructs the output media from the predicted tokens. The generative AI model is trained for media compression and decompression using a set of training data.