GAN Media Compression via Object Relevance Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI system processing of videos and images is hindered by large data volumes and resource consumption due to extraneous information, which is not effectively reduced by simple classification methods.
Innovation Solution
A system utilizing a generative adversarial network (GAN) that identifies and modifies objects in media assets based on relevance scores derived from historical data and usage contexts, reducing data volume and resource requirements by generating updated media assets recognizable to AI systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of substance
If simple classification methods are used to reduce data volume, then processing speed may be improved, but the reduction of extraneous information is not effective and data volume remains large
Solution Approach 1:
The patent segments the media asset into multiple objects using object detection, then individually processes each object based on its relevance score. This segmentation allows selective compression of less important objects while preserving critical information, effectively reducing overall data volume without sacrificing processing efficiency for important content.
Solution Approach 2:
The patent applies different compression quality levels to different regions of the media asset based on object importance. High relevance objects are preserved with high quality, while low relevance objects are compressed more aggressively. This local quality differentiation achieves effective data reduction while maintaining processing efficiency for critical elements.
2Loss of substance
If GAN generator applies aggressive modifications to reduce data volume, then compression ratio improves, but GAN discriminator may fail to identify modified objects as real
Solution Approach 1:
The patent applies partial action by selectively modifying only low relevance objects rather than all objects in the media asset. This partial modification approach achieves data volume reduction while preserving the recognizability of important objects, ensuring the GAN discriminator can still identify them as real.
Solution Approach 2:
The patent implements a feedback loop where the GAN discriminator evaluates modified objects and provides feedback to the generator. Objects that fail the discriminator's reality check are iteratively refined, ensuring that compressed representations maintain sufficient fidelity for accurate object recognition while still achieving compression.
3Device complexity
If all objects in media assets are processed equally, then processing simplicity is maintained, but resource consumption increases due to handling extraneous information
Solution Approach 1:
The patent extracts and identifies important objects from the media asset using object detection, then separates them from extraneous background elements. By taking out only the relevant objects for detailed processing and applying simplified handling to the rest, the system reduces resource consumption while maintaining manageable processing complexity.
Solution Approach 2:
The patent changes the processing parameters dynamically based on object relevance scores. High relevance objects receive intensive processing with high computational parameters, while low relevance objects are processed with reduced parameters or simplified methods. This parameter adaptation reduces overall resource consumption while maintaining processing complexity at acceptable levels.
Data Source
AI summary
An embodiment for compressing media utilizing a generative adversarial network (GAN) is provided. The embodiment may include receiving one or more media assets and historical data from a knowledge corpus in accordance with an identified usage context. The embodiment may also include identifying one or more objects in the one or more media assets. The embodiment may further include deriving a relevance score for each identified object. The embodiment may also include creating a training data set. The embodiment may further include applying one or more modifications to each object in a first set. The embodiment may also include in response to determining a GAN discriminator is able to identify each object in the first set modified by the GAN generator as real, generating one or more updated media assets including a second set of one or more objects that are identified by the GAN discriminator as real.