Feature Map Reordering for Video Coding Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing coding methods for feature map data in image processing technologies suffer from reduced coding efficiency due to discontinuity between data of different channels and high encoding bits, especially in tiling modes where channels are closely adjacent.
Innovation Solution
The proposed solution involves reordering multiple feature maps by relevance and spatially splicing them to create quasi-spatial video sequences, constructing a block-coding candidate motion information list for each sequence, and encoding the current block based on this list to generate a bitstream.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If feature maps are tiled in a single list with multiple channels closely adjacent, then the data structure is simple, but coding efficiency is reduced due to discontinuity between different channels
Solution Approach 1:
The patent segments the single list of feature maps into multiple separate lists, where each list contains feature maps of a specific channel. This segmentation allows continuous coding within each channel list while maintaining simple data structures through clear organization. The repacking module can selectively pack feature maps into appropriate lists based on channel characteristics, resolving the contradiction between structural simplicity and coding efficiency.
Solution Approach 2:
The patent introduces a new dimension of organization by creating multiple parallel lists instead of a single sequential list. This dimensional change from one-dimensional sequencing to multi-dimensional arrangement (multiple lists) enables continuous coding within each list while preserving channel-specific continuity, thereby improving coding efficiency without significantly increasing overall structural complexity.
2Loss of energy
If feature maps are reordered by relevance and spatially spliced, then compression efficiency is improved, but the encoding process becomes more complex
Solution Approach 1:
The patent applies preliminary action by performing reordering and spatial splicing of feature maps before the encoding process. The repacking module pre-organizes feature maps into optimal sequences based on channel relevance and spatial characteristics, so that subsequent encoding can proceed efficiently. This preliminary preparation reduces the overall encoding complexity while achieving better compression efficiency.
Solution Approach 2:
The repacking module acts as an intermediary between the feature map extraction and encoding processes. It introduces a intermediate processing stage that reorders and spatially splices feature maps, mediating between the raw feature maps and the encoder. This intermediary function improves compression efficiency while managing complexity by confining the complex operations to a dedicated module rather than the entire encoding system.
Data Source
AI summary
Embodiments of the disclosure provide an encoding method, a decoding method, an encoder, a decoder, and a storage medium. The method includes the following. An encoding result of a current block, a target index for the current block, and a block-coding enable flag for the current block are obtained by parsing a bitstream. A feature map splicing order is obtained. When the block-coding enable flag is valid, target block-coding motion information corresponding to the target index is determined based on the feature map splicing order and pixel position information of the current block. The encoding result of the current block is decoded based on the target block-coding motion information.