Feature Map Signaling With Presence Indicators for Selective Parsing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for signaling feature map data in neural networks are inefficient, leading to high computational complexity and resource utilization, especially in distributed systems where transmission resources are limited.
Innovation Solution
Implementing a method for encoding and decoding feature map data using region and side information presence indicators, allowing for selective parsing and setting of default values to reduce data transmission and complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If feature map data is transmitted in distributed neural network systems, then computational workload is distributed and resource utilization is improved, but transmission resources are consumed and data volume increases
Solution Approach 1:
The patent extracts only the essential presence information (presence indicators) from feature map data for transmission, rather than transmitting the complete feature map data. This allows distributed neural network processing to proceed with minimal data transmission, resolving the contradiction between distributed processing capability and data transmission volume.
Solution Approach 2:
The patent introduces presence indicators as intermediary elements that mediate between the encoder and decoder in distributed neural network systems. These indicators enable the decoder to selectively process feature map regions without receiving the complete data set, thus enabling distributed processing while minimizing transmission resources.
2Measurement precision
If complete feature map data is decoded, then processing accuracy is maintained, but computational complexity and entropy decoding time increase
Solution Approach 1:
The patent applies partial action by decoding only the regions of feature maps that are marked as present through presence indicators. The decoder selectively processes only necessary regions rather than performing complete feature map decoding, thereby maintaining processing accuracy for relevant regions while reducing overall computational complexity and entropy decoding time.
3Loss of information
If all regions of feature maps are transmitted and decoded, then information completeness is ensured, but transmission efficiency and processing speed decrease
Solution Approach 1:
The patent segments the feature map into multiple regions and applies presence indicators to mark which segments are present and need to be processed. This segmentation approach ensures that information completeness is maintained for necessary regions while eliminating transmission and processing of empty or irrelevant segments, thereby improving data processing speed without significant information loss.
Data Source
AI summary
The present disclosure relates to an efficient signaling of feature map information for a system employing a neural network. In particular, at the decoder side, a presence indicator is obtained based on information parsed from a bitstream. Based on the value of the obtained presence indicator, further data related to a feature map region are parsed or the parsing is bypassed. The presence indicator may be, for instance a region presence indicator indicating whether feature map data is included in the bitstream, or may be a side information presence indicator indicating whether a side information related to the feature map data is included in the bitstream. Similarly, an encoding method, as well as encoding and decoding devices are provided. Accordingly, the feature map data may be processed more efficiently, including reduction of decoding complexity as well as reduction of the amount of transmitted data by applying the bypassing.