Neural Network Picture Coding with Rescaling Before Downsampling

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing video compression techniques face challenges in efficiently reducing the size of video data while maintaining high-quality reconstruction, especially with the increasing demand for higher video quality and limited network resources.

Innovation Solution

A neural network framework that includes rescaling and downsampling layers to adjust input sizes before processing, minimizing unnecessary modifications and ensuring efficient compression with minimal loss of information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If video data is compressed to reduce data size for transmission or storage, then the quantity of data is decreased, but picture quality may be sacrificed

Engineering Contradiction:

Improvedata sizeVSAvoidpicture quality

Core Design Contradiction:

Quantity of substanceVSManufacturing precision

Solution Approach 1:

The patent applies rescaling before downsampling to prepare the input data in advance. By adjusting the input size to be a multiple of the downsampling ratio, the neural network can process the data more efficiently without losing important information, thus reducing data size while maintaining picture quality

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent changes the size parameter of the input data before processing. By rescaling the input to match the downsampling ratio requirements, the system optimizes the compression process to achieve better quality retention at reduced data sizes

Inventive Principle:

Principle #35Parameter changes

2Productivity

If downsampling is applied directly to reduce input size, then processing efficiency is improved, but information loss increases when input size is not a multiple of downsampling ratio

Engineering Contradiction:

Improveprocessing efficiencyVSAvoidinformation loss

Core Design Contradiction:

ProductivityVSLoss of information

Solution Approach 1:

The patent performs rescaling as a preliminary step before downsampling. This preliminary action ensures that the input size is adjusted to be a multiple of the downsampling ratio, allowing the downsampling operation to proceed efficiently without causing information loss from truncation or rounding errors

Inventive Principle:

Principle #10Preliminary action

3Device complexity

If neural network processing is applied without size adjustment, then the framework is simpler, but it cannot efficiently process various signal types with different sizes

Engineering Contradiction:

Improveframework complexityVSAvoidsignal processing adaptability

Core Design Contradiction:

Device complexityVSAdaptability or versatility

Solution Approach 1:

The patent implements a universal rescaling mechanism that can handle various signal types (images, audio, video) with different sizes. By making the framework adaptable to different input dimensions through size adjustment, the same neural network architecture can efficiently process multiple signal formats without requiring separate processing paths

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS12477131B2Method and apparatus for encoding or decoding a picture using a neural network

Publication Date: 2025.11.18 HUAWEI TECH CO LTD

AI summary

Disclosed herein are methods and systems for encoding a picture and decoding a bitstream that may represent an encoded picture. During encoding and decoding, rescaling operations are applied to rescale an input to a size that can be processed by a layer of a neural network. Embodiments disclosed herein provide methods for rescaling that achieve a reduced size of the bitstream, thereby improving compression.