Machine learning-based framework for drivable surface annotation
Enclosed are embodiments of an ML-based framework for drivable surface annotation. In an embodiment, a method comprises: obtaining, using at least one processor, multimodal map data for a geographic region; and automatically annotating, using the at least one processor, one or more semantic masks of the map data using a machine learning model.
1. A method comprising:
obtaining, using at least one processor, map data for a vehicle; and
automatically annotating, using the at least one processor, one or more semantic map layers of the map data using a machine learning model, wherein the machine learning model is trained on image samples extracted from one or more map models, and the one or more map models are split into different sampling image regions for generating a training dataset and a testing dataset, respectively.
2. The method of claim 1 , wherein the map data is multimodal.
3. The method of claim 2 , wherein the multimodal map data includes at least an intensity map and an occupancy map.
4. The method of claim 1 , wherein the machine learning model is implemented by a two-dimensional (2D) convolutional network.
5. The method of claim 1 , wherein the machine learning model is implemented using ensemble modeling.
6. The method of claim 1 , wherein the machine learning model is iteratively trained on image samples extracted from one or more map models, and different image samples are extracted for each training iteration to increase variability of training images.
7. The method of claim 6 , wherein the one or more map models are split into different sampling image regions for generating training and testing datasets for training and testing the machine learning model.
8. The method of claim 7 , wherein the one or more map models are automatically split into different sampling image regions for generating training and testing datasets using a tree data structure.
9. The method of claim 8 , wherein the one or more map models are automatically split into different sampling image regions based on a split criterion, the split criterion being when an instance of a particular road landmark is present more than once in a particular sampling image region.
10. The method of claim 8 , wherein the tree data structure is a Quadtree data structure.
11. A system comprising:
at least one processor; and
a memory storing instructions thereon that, when executed by the at least one processor, cause the at least one processor to:
obtaining map data for a vehicle; and
automatically annotating one or more semantic map layers of the map data using a machine learning model, wherein the machine learning model is trained on image samples extracted from one or more map models, and the one or more map models are split into different sampling image regions for generating a training dataset and a testing dataset, respectively.
12. The system of claim 11 , wherein the map data is localized.
13. The system of claim 11 , wherein the map data is multimodal.
14. The system of claim 13 , wherein the multimodal map data includes an intensity map and an occupancy map.
15. The system of claim 11 , wherein the machine learning model is implemented by a two-dimensional (2D) convolutional network.
16. The system of claim 11 , wherein the machine learning model is iteratively trained on image samples extracted from one or more map models, and different image samples are extracted for each training iteration to increase variability of training images.
17. The system of claim 16 , wherein the one or more map models are split into different sampling image regions for generating training and testing datasets for training and testing the machine learning model.
18. The system of claim 17 , wherein the one or more map models are automatically split into different sampling image regions for generating training and testing datasets using a tree data structure.
19. The system of claim 18 , wherein the tree data structure is a Quadtree data structure.
20. A non-transitory, computer-readable storage medium having instructions stored thereon, that when executed by at least one processor, cause the at least one processor to perform operations, comprising:
obtaining map data for a vehicle; and
automatically annotating one or more semantic map layers of the map data using a machine learning model, wherein the machine learning model is iteratively trained on image samples extracted from one or more map models, different image samples are extracted for each training iteration, and the one or more map models are automatically split into different sampling image regions for generating a training dataset and a testing dataset, respectively, using a tree data structure.