CLADE-Avg Smoothing Layer for GAN Image Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image-to-image models using Generative Adversarial Networks (GANs) face challenges in efficiently compressing images while maintaining high quality, often resulting in increased computational costs.
Innovation Solution
The introduction of a novel variant to CLADE parameters, referred to as CLADE-Avg, which adds an extra layer of average smoothing between the parameter and normalization layers, smoothing abrupt boundaries and introducing more possible values for scaling and shift, without increasing computational overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If image-to-image models use GANs for image compression, then image quality can be maintained, but computational costs increase
Solution Approach 1:
The patent modifies the normalization layer parameters by introducing average smoothing to the scaling and shift parameters, changing their computational characteristics to reduce overall computational cost while preserving image quality reconstruction capabilities
Solution Approach 2:
The patent extracts and removes unnecessary computational operations from the GAN architecture by replacing traditional normalization layers with simplified versions that use pre-computed average values, eliminating redundant calculations while maintaining essential functionality
2Productivity
If CLADE parameters are used in normalization layers, then image compression efficiency improves, but abrupt boundaries are introduced
Solution Approach 1:
The patent introduces an intermediary averaging operation between the CLADE parameter computation and the normalization layer application, which smooths abrupt boundaries by blending parameter values across spatial locations while preserving the overall compression efficiency benefits
Solution Approach 2:
The patent applies spatially-varying scaling and shift parameters that are computed through average smoothing, allowing different regions of the image to have optimized parameters while maintaining continuity and avoiding abrupt boundaries through the smoothing operation
Data Source
AI summary
System and methods for compressing image-to-image models. Generative Adversarial Networks (GANs) have achieved success in generating high-fidelity images. An image compression system and method adds a novel variant to class-dependent parameters (CLADE), referred to as CLADE-Avg, which recovers the image quality without introducing extra computational cost. An extra layer of average smoothing is performed between the parameter and normalization layers. Compared to CLADE, this image compression system and method smooths abrupt boundaries, and introduces more possible values for the scaling and shift. In addition, the kernel size for the average smoothing can be selected as a hyperparameter, such as a 3×3 kernel size. This method does not introduce extra multiplications but only addition, and thus does not introduce much computational overhead, as the division can be absorbed into the parameters after training.