Image Disparity Estimation With Cross-Attention Feature Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating disparity between two images using machine learning have limitations in accuracy.
Innovation Solution
A learning apparatus configured with a model that includes a feature generation unit and a map generation unit, utilizing a cross-attention layer to generate a disparity map based on features from two images, and a correction unit for time-series data to refine disparity estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning methods are used to estimate disparity between two images, then disparity estimation accuracy is improved, but there is still room for improvement in accuracy
Solution Approach 1:
The patent divides the image processing into multiple segments: a first image and a second image are processed separately through feature generation units to produce first features and second features. The cross-attention layer then processes these segmented features to generate disparity maps, allowing focused processing of different image regions and improving overall accuracy.
Solution Approach 2:
The cross-attention layer serves as an intermediary mechanism that receives first features from the first image and second features from the second image, processes them together, and generates the disparity map. This intermediary structure enables effective integration of information from both images to achieve accurate disparity estimation.
2Measurement precision
If a cross-attention layer is introduced to process features from two images, then pixel association accuracy is improved, but model complexity increases
Solution Approach 1:
The cross-attention layer is designed to handle multiple functions within a unified structure: it processes first features from the first image, second features from the second image, and generates disparity maps. This multi-functional design improves pixel association accuracy while managing model complexity through a single versatile component rather than multiple separate mechanisms.
Data Source
AI summary
A learning apparatus generates output data representing a disparity between first and second images in input data by inputting the input data to a model, and updates a parameter of the model to reduce a loss obtained by inputting the output data and ground truth data to a loss function. The model includes a feature generation unit configured to generate first and second features based on the first and second images, respectively, and a map generation unit configured to generate a disparity map of the disparity between the first and second images based on the first and second features. The map generation unit includes a cross-attention layer configured to receive inputs based on the first and second features. The disparity map is based on an output from the cross-attention layer.