Compression Neural Network Padding Sequence for Higher Efficiency

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional data compression networks increase the padding area as the ratio of downscaling/upscaling increases, deteriorating compression efficiency.

Innovation Solution

Implement multiple padding and downscaling operations in an encoding apparatus, followed by corresponding upscaling and trimming operations in a decoding apparatus, to manage image size adjustments efficiently.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If the ratio of downscaling/upscaling in the data compression network is increased, then compression efficiency is improved, but the padding area is further increased

Engineering Contradiction:

Improvecompression efficiencyVSAvoidpadding area

Core Design Contradiction:

ProductivityVSArea of stationary object

Solution Approach 1:

The patent segments the single padding operation into multiple smaller padding operations performed at different stages of the compression network. Instead of applying all padding at once before downscaling, the network applies padding progressively at multiple intermediate stages, reducing the total padding area required while maintaining the ability to handle various input sizes and achieving high compression efficiency.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces a temporal dimension to the padding operation by distributing it across multiple time steps or stages in the compression network. This transforms a spatial problem (padding area in 2D image space) into a temporal process where padding is applied incrementally, allowing the network to achieve high compression ratios without proportionally increasing the total padding area.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

2Adaptability or versatility

If multiple padding operations are applied to handle various input sizes, then adaptability is improved, but device complexity is increased

Engineering Contradiction:

Improvehandling various input sizesVSAvoidcomplexity of multiple padding/downscaling operations

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent designs the multiple padding and downscaling operations to serve universal functions throughout the compression network. Each stage uses the same operational pattern (padding followed by downscaling), allowing the network to handle various input sizes through a standardized multi-stage process rather than requiring different architectures for different input dimensions, thus improving adaptability without proportionally increasing complexity.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS20250371654A1Method, apparatus, and storage medium using padding/trimming in compression neural network

Publication Date: 2025.12.04 ELECTRONICS & TELECOMM RES INST

AI summary

An encoding apparatus extracts features of an image by applying multiple padding operations and multiple downscaling operations to an image represented by data and transmits feature information indicating the features to a decoding apparatus. The multiple padding operations and the multiple downscaling operations are applied to the image in an order in which one padding operation is applied and thereafter one downscaling operation corresponding to the padding operation is applied. A decoding method receives feature information from an encoding apparatus, and generates a reconstructed image by applying multiple upscaling operations and multiple trimming operations to an image represented by the feature information. The multiple upscaling operations and the multiple trimming operations are applied to the image in an order in which one upscaling operation is applied and thereafter one trimming operation corresponding to the upscaling operation is applied.