Method and system for a fast adaptation for image segmentation for autonomous edge vehicles
A method includes obtaining, by a local data system manager of a local data system of the local data systems, a portion of unlabeled data from a local data source, performing, using a domain classifier in the local data system manager, a domain classification analysis on the portion of the unlabeled data to identify a domain of the unlabeled data, making a first determination, based on the domain classification, that the domain classification has significantly varied from a previous domain, based on the first determination: performing an adaptive procedure on a local data system image segmentation model to obtain an adapted image segmentation model, and performing a domain reclassification on the domain classifier to obtain an updated domain classifier, and implementing the adapted image segmentation model on the local data system.
1. A method for managing local data systems, the method comprising:
obtaining, by a local data system manager of a local data system of the local data systems, a portion of unlabeled data from a local data source;
performing, using a domain classifier in the local data system manager, a domain classification analysis on the portion of the unlabeled data to identify a domain of the unlabeled data;
making a first determination, based on the domain classification, that the domain classification has significantly varied from a previous domain;
based on the first determination:
performing an adaptive procedure on a local data system image segmentation model to obtain an adapted image segmentation model; and
performing a domain reclassification on the domain classifier to obtain an updated domain classifier; and
implementing the adapted image segmentation model on the local data system.
2. The method of claim 1 , wherein the local data system is an autonomous vehicle edge device.
3. The method of claim 1 , wherein the portion of the unlabeled data is a set of unlabeled images.
4. The method of claim 1 , wherein the local data system image segmentation model is obtained from an image segmentation manager.
5. The method of claim 4 , wherein the image segmentation manager is a cloud-based service.
6. The method of claim 4 , wherein performing the adaptive procedure comprises:
obtaining a portion of a labeled training dataset from the image segmentation manager;
dividing the portion of the labeled training dataset to obtain groups of labeled data;
dividing the portion of unlabeled data to obtain groups of unlabeled data;
implementing a machine learning algorithm on the groups of labeled data and the groups of unlabeled data using the local data system image segmentation model to obtain the adapted image segmentation model.
7. The method of claim 6 ,
wherein the machine learning algorithm is a domain adaptation neural network (DANN),
wherein the local data system image segmentation model is generated using a convolutional neural network (CNN), and
wherein the DANN comprises:
determining a decay ratio; and
optimizing a number of layers in an extraction phase of a neural network using the decay ratio.
8. A non-transitory computer readable medium comprising computer readable program code, which when executed by a computer processor enables the computer processor to perform a method for managing local data systems, the method comprising:
obtaining, by a local data system manager of a local data system of the local data systems, a portion of unlabeled data from a local data source;
performing, using a domain classifier in the local data system manager, a domain classification analysis on the portion of the unlabeled data to identify a domain of the unlabeled data;
making a first determination, based on the domain classification, that the domain classification has significantly varied from a previous domain;
based on the first determination:
performing an adaptive procedure on a local data system image segmentation model to obtain an adapted image segmentation model; and
performing a domain reclassification on the domain classifier to obtain an updated domain classifier; and
implementing the adapted image segmentation model on the local data system.
9. The non-transitory computer readable medium of claim 8 , wherein the local data system is an autonomous vehicle edge device.
10. The non-transitory computer readable medium of claim 8 , wherein the portion of the unlabeled data is a set of unlabeled images.
11. The non-transitory computer readable medium of claim 8 , wherein the local data system image segmentation model is obtained from an image segmentation manager.
12. The non-transitory computer readable medium of claim 11 , wherein the image segmentation manager is a cloud-based service.
13. The non-transitory computer readable medium of claim 11 , wherein performing the adaptive procedure comprises:
obtaining a portion of a labeled training dataset from the image segmentation manager;
dividing the portion of the labeled training dataset to obtain groups of labeled data;
dividing the portion of unlabeled data to obtain groups of unlabeled data; and
implementing a machine learning algorithm on the groups of labeled data and the groups of unlabeled data using the local data system image segmentation model to obtain the adapted image segmentation model.
14. The non-transitory computer readable medium of claim 13 ,
wherein the machine learning algorithm is a domain adaptation neural network (DANN),
wherein the local data system image segmentation model is generated using a convolutional neural network (CNN), and
wherein the DANN comprises:
determining a decay ratio; and
optimizing a number of layers in an extraction phase of a neural network using the decay ratio.
15. A system comprising:
an image segmentation manager comprising circuitry;
a local data system comprising a processor; and
memory including instructions, which when executed by the processor, perform a method comprising:
obtaining, by a local data system manager of the local data system, a portion of unlabeled data from a local data source;
performing, using a domain classifier in the local data system manager, a domain classification analysis on the portion of the unlabeled data to identify a domain of the unlabeled data;
making a first determination, based on the domain classification, that the domain classification has significantly varied from a previous domain;
based on the first determination:
performing an adaptive procedure on a local data system image segmentation model to obtain an adapted image segmentation model; and
performing a domain reclassification on the domain classifier to obtain an updated domain classifier; and
implementing the adapted image segmentation model on the local data system.
16. The system of claim 15 , wherein the local data system is an autonomous vehicle edge device.
17. The system of claim 15 , wherein the portion of the unlabeled data is a set of unlabeled images.
18. The system of claim 15 , wherein the local data system image segmentation model is obtained from the image segmentation manager.
19. The system of claim 18 , wherein the image segmentation manager is a cloud-based service.
20. The system of claim 18 , wherein performing the adaptive procedure comprises:
obtaining a portion of a labeled training dataset from the image segmentation manager;
dividing the portion of the labeled training dataset to obtain groups of labeled data;
dividing the portion of unlabeled data to obtain groups of unlabeled data; and
implementing a machine learning algorithm on the groups of labeled data and the groups of unlabeled data using the local data system image segmentation model to obtain the adapted image segmentation model.