Image Translation Learning With Masking for Robust Unsupervised Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing unsupervised methods for image representation learning, such as predicting image rotation and geometric transformations, struggle with rotation-invariant images and edge effects, leading to poor representation learning performance.
Innovation Solution
A method and system for unsupervised deep representation learning based on image translation, involving a random translation transformation, an image mask module, a deep neural network, a regression loss function, and a feature extraction module, to eliminate edge effects and focus on salient objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If image rotation prediction is used for unsupervised representation learning, then the method can handle general images, but it fails to learn meaningful features from rotation-invariant images
Solution Approach 1:
The patent changes the transformation parameter from rotation angle to translation distance. By using translation transformations with varying distances instead of fixed rotation angles, the method can effectively process both rotation-sensitive and rotation-invariant images, resolving the contradiction between versatility and representation quality.
2Adaptability or versatility
If geometric transformations including translation are used, then rotation insufficiency is compensated, but edge effects are introduced that simplify the prediction task and reduce learning effectiveness
Solution Approach 1:
The patent applies different masking strategies to different regions of the image. By masking out edge regions where translation artifacts occur and preserving only the central salient objects, the method eliminates the harmful edge effects while maintaining the usefulness of translation transformations for representation learning.
3Adaptability or versatility
If 72 geometric transformations are applied to each image, then transformation diversity is increased, but the prediction task becomes too simple due to edge effects and learning effectiveness is reduced
Solution Approach 1:
The patent extracts and removes the harmful edge regions from the transformed images through masking. By taking out the edge effects that make prediction too simple and retaining only the meaningful central regions, the method maintains transformation diversity while improving representation learning effectiveness.
Data Source
AI summary
A system for unsupervised deep representation learning based on image translation is provided. The system includes an image translation transformation module used for performing a random translation transformation on an image and generating an auxiliary label; an image mask module connected with the image translation transformation module and used for applying a mask to the image after translation transformation; a deep neural network connected with the image mask module and used for predicting an actual auxiliary label of the image after the mask is applied and learning the deep representation of the image; a regression loss function module connected with the deep neural network and used for updating parameters of the deep neural network based on a loss function; and a feature extraction module connected with the deep neural network and used for extracting the representation of the image.