Content-Adaptive Online Block Training for Neural Image Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural network-based video or image coding frameworks are limited in accommodating various types of compression frameworks, leading to increased computing memory and cost, as well as rate-distortion loss, resulting in lower overall performance.
Innovation Solution
The method involves content-adaptive online training for multiple blocks in neural image compression, where an input image is split into blocks, a neural network is fine-tuned based on these blocks, updated parameters are computed, and the network is updated and used to generate a compressed representation of the blocks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional systems use increased computing memory and cost to accommodate various types of compression frameworks, then the system can support more framework types, but the overall performance of the image or video framework deteriorates
Solution Approach 1:
The patent implements a universal neural network architecture that can handle multiple compression frameworks through a unified end-to-end training approach. The system uses a single neural network model that performs both encoding and decoding functions, eliminating the need for separate framework-specific implementations while maintaining high performance across different compression scenarios
2Adaptability or versatility
If conventional systems use increased computing memory, then various types of frameworks can be accommodated, but rate-distortion loss increases
Solution Approach 1:
The patent dynamically adjusts neural network parameters including learning rates, batch sizes, and compression ratios based on the specific framework being used. This adaptive parameter tuning allows the system to optimize rate-distortion performance for each framework type without requiring increased memory resources, achieving low loss through intelligent parameter configuration rather than brute-force resource allocation
3Device complexity
If a single neural network is used for multiple compression frameworks, then device complexity is reduced, but the system must process larger data blocks
Solution Approach 1:
The patent divides the neural network processing into multiple stages including feature extraction, compression, and reconstruction modules. Each stage processes smaller intermediate data representations rather than requiring the entire image to be processed as one large block, thereby reducing memory requirements while maintaining the benefits of a unified neural network architecture
Data Source
AI summary
Content-adaptive online training for end-to-end (E2E) neural image compression (NIC) using a neural network performed by at least one processor, is provided, including receiving an input image, to an E2E NIC framework, including one or more blocks, preprocessing a first neural network of the E2E NIC framework, based on the one or more blocks, computing updated parameters using the preprocessed first neural network, encoding the one or more blocks and the updated parameters, updating the first neural network based on the encoded updated parameters, and generating a compressed representation of the encoded one or more blocks using the updated first neural network.