Stylized Image Generation via Model Parameter Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing algorithms require a large amount of data and training samples to construct effective style transfer models, which is costly and inefficient, especially when samples of specific styles are scarce.
Innovation Solution
A method and apparatus for generating stylized images by constructing a target style data generation model using transferred model parameters from a face image generation model, without requiring a large number of training samples. This involves training two sample generation models based on different style types and fusing their model parameters to create a stylized image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large amount of training samples are used to construct an image processing algorithm model, then the model effectiveness is improved, but the construction cost increases significantly
Solution Approach 1:
The patent transforms the training process by changing parameters from requiring numerous style-specific samples to using a pre-trained face generation model with style parameters. The style information is encoded as parameters rather than requiring extensive sample training, thus reducing the quantity of training data needed while maintaining model effectiveness.
Solution Approach 2:
The patent applies preliminary action by pre-training a face image generation model before style transfer. The base model is already trained on large datasets, and style information is subsequently integrated through parameter adjustment rather than retraining from scratch, reducing the need for additional training samples.
2Quantity of substance
If traditional style transfer models are constructed without sufficient style-specific training data, then the construction cost is reduced, but the model cannot be effectively constructed for that style type
Solution Approach 1:
The patent achieves universality by creating a model that can handle multiple style types through a unified framework. The pre-trained face generation model serves as a universal base that can be adapted to different styles by adjusting style parameters, eliminating the need for separate models for each style type.
Solution Approach 2:
The patent uses copying by replicating the successful architecture and training approach of face generation models and adapting them for style transfer. By copying the proven effectiveness of face generation models and applying it to style transfer through parameter adjustment, the method achieves model constructability without extensive style-specific data.
3Adaptability or versatility
If two style types are fused using traditional methods, then a comprehensive stylized image is generated, but a large number of fused training samples are required
Solution Approach 1:
The patent applies parameter changes by representing style information as adjustable parameters rather than requiring fused training samples. Style fusion is achieved by modifying model parameters to combine characteristics of different styles, eliminating the need for extensive fused sample training while maintaining adaptability to multiple style combinations.
Data Source
AI summary
Embodiments of the present disclosure provide a method and apparatus for generating a stylized image, an electronic device and a storage medium. The method includes: acquiring model parameters to be transferred of a face image generation model to construct a first sample generation model to be trained and a second sample generation model to be trained; respectively training corresponding sample generation models to be trained based on training samples of a first style type and a training sample of a second style type to obtain a first target sample generation model and a second target sample generation model; and determining a target style data generation model based on model parameters to be fitted of the two target sample generation models to generate, based on the target style data generation model, a stylized image in which the two style types are fused.