Semi-supervised Makeup Transfer Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional makeup transfer applications using unsupervised learning models face inefficiencies due to the need for aligned images, instability in network training, and unsatisfactory color transfer results, while supervised learning models require extensive paired data and are time-consuming to train, limiting user control and selectivity in makeup transfer.
Innovation Solution
A semi-supervised learning approach is employed, where the neural network is trained using a combination of supervised learning with a small amount of paired data and unsupervised learning with unpaired data, allowing for better initialization and stabilization, enabling more efficient and effective makeup transfer with improved color accuracy and user control over reference images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If unsupervised learning models are used for makeup transfer, then training is easier and user control is improved, but color transfer efficiency and makeup transfer quality deteriorate
Solution Approach 1:
The patent combines supervised learning and unsupervised learning approaches into a hybrid model. The supervised learning component ensures accurate color transfer and high-quality makeup transfer results, while the unsupervised learning component maintains ease of training and user control. This merging resolves the contradiction by integrating the strengths of both approaches.
2Manufacturing precision
If supervised learning models are used for makeup transfer, then makeup transfer quality and color accuracy are improved, but training time and data preparation requirements increase
Solution Approach 1:
The patent applies partial supervised learning by using only a small subset of paired data for training, rather than requiring extensive paired datasets. This partial application of supervised learning maintains adequate makeup transfer quality while significantly reducing training time and data preparation requirements.
3Measurement precision
If aligned images are required for training, then training precision is improved, but adaptability to different poses and user control deteriorate
Solution Approach 1:
The patent segments the training data into aligned and unaligned subsets, using both types of images for training. This segmentation allows the model to learn from precisely aligned images while also adapting to various poses and orientations, resolving the contradiction between training precision and pose adaptability.
Data Source
AI summary
The present disclosure relates to systems, computer-implemented methods, and non-transitory computer readable medium for automatically transferring makeup from a reference face image to a target face image using a neural network trained using semi-supervised learning. For example, the disclosed systems can receive, at a neural network, a target face image and a reference face image, where the target face image is selected by a user via a graphical user interface (GUI) and the reference face image has makeup. The systems transfer, by the neural network, the makeup from the reference face image to the target face image, where the neural network is trained to transfer the makeup from the reference face image to the target face image using semi-supervised learning. The systems output for display the makeup on the target face image.