Identity Swapping Model Training with Fake Template Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The lack of real labeled images in the training process of identity swapping models makes the training process uncontrollable, resulting in poor quality identity swapping images.
Innovation Solution
The introduction of fake template and labeled sample groups, where a real labeled image is used to constrain the training process, and a fake labeled image is generated by performing identity swapping on a real template image, allowing for a more controlled training of the identity swapping model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If unsupervised training procedure is used for identity swapping model, then training process can be performed without real labeled images, but the training process becomes uncontrollable and generates poor quality identity swapping images
Solution Approach 1:
The patent applies preliminary action by pre-processing images to generate fake template images and fake labeled images before the actual training process. These pre-processed images serve as controlled training data that enables the model to learn identity swapping while maintaining controllability and image quality, resolving the contradiction between ease of training and image quality.
Solution Approach 2:
The patent introduces fake template images and fake labeled images as intermediary elements between the source image and template image. These intermediary images act as a bridge that allows the training process to be both easy to perform (without requiring real labeled identity swapping images) and maintain high quality output through controlled training data.
2Manufacturing precision
If real labeled images are used to constrain the training process, then the training becomes controllable and image quality improves, but the complexity of data preparation increases
Solution Approach 1:
The patent applies self-service by enabling the system to automatically generate its own training data (fake template images and fake labeled images) from available source images and template images without requiring external real labeled identity swapping images. This automated data generation process reduces the complexity of manual data preparation while maintaining training controllability and image quality.
Solution Approach 2:
The patent changes the parameters of training data by transforming real images into fake template images and fake labeled images through specific processing operations. This parameter transformation allows the system to create controlled training data with desired properties, improving image quality while automating the data preparation process and reducing overall complexity.
3Manufacturing precision
If fake template images and fake labeled images are generated and used for training, then training controllability and image quality improve, but the processing time and computational resources increase
Solution Approach 1:
The patent applies partial action by generating fake template images and fake labeled images selectively for the specific purpose of training the identity swapping model, rather than processing all possible images. This targeted approach maintains high training controllability and image quality while limiting the additional processing time to only the necessary subset of images required for effective training.
Data Source
AI summary
An image processing method including obtaining a fake template sample group comprising a first source image, a real labeled image, and a fake template image, inputting the fake template image into an identity swapping model to obtain a first identity swapping image of the fake template image, obtaining a fake labeled sample group comprising a second source image, a real template image, and a fake labeled image, the fake labeled image being based on identity swapping processing of the real template image, inputting the real template image into the identity swapping model to obtain a second identity swapping image of the real template image, and training the identity swapping model based on the fake template sample group, the first identity swapping image, the fake labeled sample group, and the second identity swapping image to generate a trained identity swapping model.