CNN Feature Decoding Using Metadata-Based Split Layer Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression technologies for machine vision tasks face challenges in efficiently encoding and decoding tensors from convolutional neural networks, particularly in identifying and utilizing the appropriate starting layers for feature coding, which affects the effectiveness of video compression and processing.
Innovation Solution
A method and system for encoding and decoding tensors from convolutional neural networks by determining a starting layer of a second part of the neural network based on decoded information, allowing for efficient splitting of the CNN into a backbone and head network, and using metadata to identify the CNN architecture and split points for accurate decoding and encoding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the CNN is split into a backbone network and a head network for feature coding, then the video compression efficiency is improved, but the complexity of identifying the correct starting layer increases
Solution Approach 1:
The patent stores metadata indicating the starting layer of the head network in advance during the model training phase. This preliminary action allows the decoding device to directly retrieve the starting layer information without performing complex analysis during video compression, thus resolving the contradiction between compression efficiency and identification complexity
Solution Approach 2:
The patent introduces metadata as an intermediary element that carries the starting layer information between the encoding and decoding processes. This metadata acts as a bridge that simplifies the interaction between the backbone network and head network, enabling efficient feature coding without increasing system complexity
2Ease of operation
If the starting layer of the head network is not accurately identified, then the encoding and decoding process is simplified, but the compatibility between backbone network and head network deteriorates
Solution Approach 1:
The starting layer information is determined and stored in advance during model training. This preliminary determination ensures that the correct starting layer is identified before the encoding process begins, maintaining compatibility between the backbone network and head network while keeping the actual encoding and decoding operations simple
Solution Approach 2:
The patent uses metadata that provides feedback information about the starting layer configuration. This feedback mechanism ensures that the decoding device can accurately identify the starting layer and maintain proper compatibility with the backbone network, while the encoding process remains straightforward
Data Source
AI summary
A system and method of decoding information for data generated by a first part of a neural network. The method comprises decoding information for determining at least a starting layer of a second part of the neural network, the neural network including at least the first part and the second part, the second part being different from the first part; and determining the starting layer of the second part of the neural network based on the decoded information.