Neural Network Sub-Networks Rescaling for Video Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video compression techniques face challenges in achieving high compression ratios with minimal loss in picture quality, especially when dealing with limited network resources and increasing demands for higher video quality.
Innovation Solution
A neural network framework comprising at least two sub-networks, where one sub-network applies downsampling to the input picture after rescaling it to an integer multiple of the combined downsampling ratio, resulting in a reduced bitstream size while ensuring minimal information loss during decoding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If downsampling is applied to reduce bitstream size, then compression ratio is improved, but picture quality deteriorates due to information loss
Solution Approach 1:
The patent applies rescaling to the input picture before downsampling to ensure the dimensions are compatible with the downsampling operation. This preliminary action prevents information loss by properly preparing the input data structure, allowing the downsampling to proceed without discarding valuable picture information while still achieving compression.
Solution Approach 2:
The patent divides the downsampling process into multiple sequential downsampling layers, each with a specific downsampling ratio. This segmentation allows for controlled compression at different stages, maintaining picture quality by processing the image in manageable steps rather than applying a single aggressive downsampling operation.
2Quantity of substance
If multiple downsampling layers are used to achieve higher compression, then bitstream size is reduced, but device complexity increases
Solution Approach 1:
The patent employs a single neural network that performs multiple functions: it handles rescaling, applies multiple downsampling layers with different ratios, and manages the entire compression process. This multi-functional approach reduces device complexity by consolidating what could be separate processing stages into one unified network, achieving high compression without proportionally increasing system complexity.
3Loss of information
If rescaling is applied before downsampling to maintain picture quality, then information loss is reduced, but processing time increases
Solution Approach 1:
The patent integrates rescaling and downsampling into a continuous processing flow within the neural network, where the rescaling operation seamlessly transitions into the downsampling layers. This continuous action eliminates the need for separate processing steps, maintaining picture quality through proper rescaling while minimizing processing time by avoiding interruptions or redundant operations between stages.
Data Source
AI summary
A method for encoding a picture and decoding a bitstream that represents a picture using a neural network (NN) that comprises a plurality of sub-networks is provided. The method includes applying, before processing an input with the at least one sub-network comprising at least two downsampling layers, a rescaling to the input, wherein the rescaling comprises changing the size S1 in the at least one dimension to be S1 so that S1 is an integer multiple of a combined downsampling ratio Rk of the at least one sub-network, after the rescaling, processing the input by the at least one sub-network comprising at least two downsampling layers and providing an output with the size S2, wherein S2 is smaller than S1, and providing, after processing the picture using the NN, a bitstream as output.