Neural Feature Map Block Alignment for Better Compression

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing encoders for feature maps in Faster-RCNN systems suffer from poor compression efficiency due to the arrangement of feature maps without considering spatial correlation, leading to inefficient data transmission and decoding.

Innovation Solution

Generate unit feature maps by packing pixels into encoded blocks and aligning the upper left boundaries of these maps with encoded blocks, allowing for improved compression efficiency through optimized block arrangements and encoding strategies.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If feature maps are arranged in order from upper left without boundary alignment, then the arrangement process is simple, but compression efficiency deteriorates

Engineering Contradiction:

Improvearrangement process simplicityVSAvoidcompression efficiency

Core Design Contradiction:

Ease of manufactureVSLoss of energy

Solution Approach 1:

The feature maps are divided into multiple encoded blocks with aligned boundaries. Each encoded block corresponds to a specific region of feature maps, allowing independent processing and optimization of compression for each block while maintaining overall efficiency

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces a new dimension of boundary alignment in the spatial arrangement of feature maps. By aligning encoded block boundaries with feature map boundaries in the horizontal and vertical directions, it creates an optimized two-dimensional layout that improves compression efficiency without significantly complicating the arrangement process

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

2Ease of operation

If feature maps are arranged without boundary alignment, then encoding process is straightforward, but compression efficiency deteriorates

Engineering Contradiction:

Improveencoding process simplicityVSAvoidcompression efficiency

Core Design Contradiction:

Ease of operationVSLoss of energy

Solution Approach 1:

The patent performs preliminary boundary alignment between encoded blocks and feature maps before the encoding process. By pre-organizing the feature maps so that their boundaries align with encoded block boundaries, it simplifies the subsequent encoding operation while achieving superior compression efficiency

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent changes the spatial parameters (position and size) of encoded blocks to match the dimensions and locations of feature maps. This parameter optimization ensures that each encoded block efficiently contains complete feature map regions, improving compression without complicating the encoding workflow

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20260019579A1Encoding device, decoding device, encoding method, and decoding method

Publication Date: 2026.01.15 PANASONIC INTELLECTUAL PROPERTY CORP OF AMERICA

AI summary

This encoder comprises circuitry and a memory connected to the circuitry. The circuitry generates a plurality of feature maps by means of a neural network having one or more layers on the basis of an input image to be processed, generates a plurality of unit feature maps on the basis of the plurality of feature maps by packing a plurality of pixels included in at least one feature map into at least one encoded block, generates a picture by arranging a plurality of encoded blocks corresponding to the plurality of unit feature maps, and encodes the picture into a bitstream, and in the generation of the unit feature maps, the upper left boundary of each unit feature map is matched with the upper left boundary of any encoded block.