Feature Map Encoding via Transform Vector Entropy Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Video Coding for Machines (VCM) technologies face inefficiencies in encoding and decoding feature maps, leading to excessive data transmission and reduced compression performance.
Innovation Solution
The method involves generating multiple feature maps using an input image, transforming them with a transform vector, and creating a bitstream through entropy encoding, while also reconstructing feature maps from the bitstream using entropy decoding and inverse transformation, optimizing the process with artificial neural networks and transform unit groups.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If feature maps are transmitted directly without transformation, then machine task performance is maintained, but data transmission amount is excessive
Solution Approach 1:
The patent extracts only the essential transform coefficients and transform vectors from the complete feature maps, transmitting only these critical components rather than the entire feature map data. This extraction approach maintains the necessary information for machine tasks while significantly reducing transmission volume.
Solution Approach 2:
The patent transforms feature maps into a different parameter space using transform vectors, changing the representation from spatial domain to frequency domain coefficients. This parameter transformation enables more efficient compression while preserving the essential characteristics needed for machine task performance.
2Quantity of substance
If transform vectors are used to compress feature maps, then data transmission is reduced, but decoding complexity increases
Solution Approach 1:
The patent performs preliminary organization and packing of transform coefficients and vectors into structured bitstream formats during encoding. This preliminary structuring simplifies the decoding process by pre-arranging data in a manner that requires minimal reorganization during reconstruction, thereby reducing decoding complexity despite the transformation overhead.
3Measurement precision
If multiple feature maps are generated and transmitted, then machine task accuracy is improved, but transmission bandwidth is consumed
Solution Approach 1:
The patent merges multiple feature maps into a unified transform domain representation, combining their essential coefficients and shared transform vectors into a single compressed bitstream. This merging approach preserves the information from multiple feature maps needed for accurate machine tasks while transmitting them as one integrated data structure, reducing overall bandwidth consumption.
Data Source
AI summary
Disclosed herein is an encoding method. The encoding method includes generating multiple feature maps using an input image, transforming the feature maps using a transform vector, and generating a bitstream by performing entropy encoding on at least any one of the feature map, the transform coefficient of the feature map, or the transform vector, or a combination thereof.