Decoder Feature Map Generation Method for Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The large data amount of feature maps generated in Faster-RCNN increases the data transmission burden between an encoder and a decoder, leading to inefficiencies in object detection systems.
Innovation Solution
A decoder dynamically selects between methods to generate feature maps using intermediate feature maps or alternative feature maps from different hierarchical layers, reducing data transmission by omitting unnecessary encoding and enabling efficient reconstruction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all feature maps are encoded and transmitted from encoder to decoder, then inference accuracy is maintained, but data transmission amount increases
Solution Approach 1:
The patent extracts and transmits only the necessary feature map information from the encoder to the decoder. By selectively encoding only certain feature maps or their essential representations rather than all feature maps, the system reduces data transmission volume while preserving the information needed for accurate object detection inference.
Solution Approach 2:
The patent creates a compressed representation or copy of the essential feature map information that can be transmitted efficiently. Instead of transmitting the full feature maps, the system encodes and transmits a reduced representation that maintains the necessary semantic information for the decoder to perform accurate inference.
2Productivity
If feature maps are generated for all hierarchical layers, then detection completeness is improved, but processing complexity increases
Solution Approach 1:
The patent segments the feature map generation process by hierarchical layers and selectively processes only the necessary layers. Instead of uniformly processing all hierarchical layers, the system identifies and processes only the critical layers needed for detection, reducing overall processing complexity while maintaining detection completeness.
Solution Approach 2:
The patent applies different processing qualities to different hierarchical layers based on their importance. Critical layers receive full processing and encoding, while less critical layers are processed with reduced complexity or omitted entirely, optimizing the balance between detection completeness and processing complexity.
Data Source
AI summary
A decoder includes: circuitry; and a memory connected to the circuitry, in which the circuitry, in operation, generates an image by decoding a bitstream, generates a plurality of intermediate feature maps having a uniform size based on the image, generates, based on the plurality of intermediate feature maps, a plurality of feature maps having different sizes for each hierarchical layer in a plurality of hierarchical layers included in a neural network for a machine task, and in generation of the feature maps, for at least one hierarchical layer of the plurality of hierarchical layers, selects any of: a first method of generating the plurality of feature maps of the at least one hierarchical layer using the plurality of intermediate feature maps, and a second method of generating the plurality of feature maps of the at least one hierarchical layer without using the plurality of intermediate feature maps but using the plurality of feature maps generated in a hierarchical layer different from the at least one hierarchical layer of the plurality of hierarchical layers.