Uncertainty-refined image segmentation under domain shift
| DWPI Title: Method for digital image segmentation of image data, involves replacing segmentation label of any image element with uncertainty value above predefined threshold with new-segmentation label corresponding to segmentation class |
| Abstract: A method for digital image segmentation is provided. The method comprises training a neural network for image segmentation with a labeled training dataset from a first domain, wherein a subset of nodes in the neural net are dropped out during training. The neural network receives image data from a second, different domain. A vector of N values that sum to 1 is calculated for each image element, wherein each value represents an image segmentation class. A label is assigned to each image element according to the class with the highest value in the vector. Multiple inferences are performed with active dropout layers for each image element, and an uncertainty value is generated for each image element. The label of any image element with an uncertainty value above a predefined threshold is replaced with a new label corresponding to the class with the next highest value. |
| Use: Method for digital image segmentation of image data. |
| Advantage: The method enables utilizing a cloud computing as a model of service delivery for enabling convenient, on-demand network access to a shared pool of configurable computing resources that can be rapidly provisioned and released with minimal management effort or interaction with a provider of the service. The method allows a cloud consumer to unilaterally provision computing capabilities, such as server time and network storage, as needed automatically without requiring human interaction with the service's provider. The cloud computing environment is allowed to offer infrastructure, platforms and/or software as services for which the cloud consumer does not need to maintain resources on a local computing device. The capabilities are rapidly and elastically provisioned, in some cases automatically, to quickly scale out and quickly scale in. |
| Novelty: The digital image segmentation method involves training (502) a neural network for image segmentation with a labeled training dataset from a domain, where a subset of nodes in the neural net are dropped out during training. The image data is received (504) from another domain by the neural network, where the image data comprises a number of image elements. A vector of N values that sum to 1 for each image element is calculated (506), where each of the N values represents an image-segmentation class. A segmentation label is assigned (508) to the image element. Multiple inferences for the image elements are performed (510) with active dropout layers. An uncertainty value for image elements is generated (512) according to the inferences. |
| Filed: 5/29/2020 |
| Application Number: US16887311A |
| Tech ID: SD 15006.0 |
| This invention was made with Government support under Contract No. DE-NA0003525 awarded by the United States Department of Energy/National Nuclear Security Administration. The Government has certain rights in the invention. |
| Data from Derwent World Patents Index, provided by Clarivate All rights reserved. Republication or redistribution of Clarivate content, including by framing or similar means, is prohibited without the prior written consent of Clarivate. Clarivate and its logo, as well as all other trademarks used herein are trademarks of their respective owners and used under license. |