Image Stylization System Balancing Strength and Identity Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image stylization solutions fail to produce images with high stylization strength and high identity preservation, and are inadequate in varying stylization strength and identity preservation levels to accommodate specific user requirements and use cases.
Innovation Solution
An image stylization system that refines image generation and translation pipelines through iterative training and custom loss functions, allowing for direct and feature-level image combination to achieve desired balances of stylization strength and identity preservation, using techniques such as image dataset construction, image generation models, and image translation models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Strength
If image stylization is applied to transform images from source domain to target domain, then stylization strength is improved, but identity preservation deteriorates
Solution Approach 1:
The patent segments the image transformation process into multiple stages: identity feature extraction from source images, style feature extraction from target images, and controlled combination of these features. This segmentation allows independent optimization of identity preservation and stylization strength, resolving the contradiction between maintaining original identity and achieving strong stylization effects.
Solution Approach 2:
The patent introduces controllable parameters that allow dynamic adjustment of the balance between identity preservation and stylization strength. By changing these parameters, users can optimize the transformation process for specific use cases, achieving high stylization strength when needed while maintaining identity preservation when required.
2Strength
If image stylization transforms features heavily to achieve target domain characteristics, then stylization strength is improved, but recognition of original entity deteriorates
Solution Approach 1:
The patent performs preliminary extraction and preservation of identity-critical features from source images before applying stylization transformations. By identifying and protecting these key features in advance, the system ensures that entity recognition accuracy is maintained even as stylization strength increases through subsequent transformations.
3Ease of manufacture
If existing stylization solutions apply uniform transformation to all images, then processing simplicity is maintained, but adaptability to specific user requirements deteriorates
Solution Approach 1:
The patent implements a dynamic transformation system where processing parameters, feature weights, and transformation intensity can be adjusted based on specific user requirements and image characteristics. This dynamic approach maintains processing simplicity through automated parameter selection while achieving high adaptability to different use cases and user preferences.
Data Source
AI summary
A computer-implemented method and system that constructs a set of target domain images, trains an image generation model using this set, uses the trained model to generate paired images such as target domain images paired with source domain images, evaluates a quality of the paired image set, constructs an adjusted paired image set based on the evaluated quality, and generates output target domain images using an image translation model trained on the adjusted set. A computer-implemented method and system that constructs an augmented set of target domain images including condition labels, uses it to train a conditional image producing model, generates two feature maps at a layer of the trained image producing label by using two input sets including two conditional labels, and uses the feature maps and mask to compute a combined feature map subsequently used to generate output target domain images by the trained image producing model.