Virtually boosted training
A method for training a machine learning model includes receiving real data comprising a real element in a real environment. The training also includes annotating the real element with a first annotation based on predicted attributes of the real element. The first annotation having a first format. The training further includes converting the first format of the first annotation to a second format corresponding to a ground truth annotation of the real element. The training still further includes adjusting parameters of the machine learning model to minimize a difference between values of the ground truth annotation of the real element and the converted first annotation.
1. A method for training a machine learning model, comprising:
receiving real data comprising a real element in a real environment;
annotating the real element with a first annotation based on predicted attributes of the real element, the first annotation having a first format;
determining a second format based on an existing format of a ground truth annotation of the real element, the first format and the second format representing different formats of a same annotation type;
converting the first format of the first annotation to the determined second format; and
adjusting parameters of the machine learning model to minimize a difference between values of the ground truth annotation of the real element and the converted first annotation.
2. The method of claim 1 , further comprising:
receiving simulated data comprising a simulated element in a simulated environment; and
annotating the real element with a second annotation based on predicted attributes of the simulated element, the second annotation having the first format.
3. The method of claim 2 , further comprising adjusting the parameters to minimize a difference between values of a ground truth annotation of the simulated element and the second annotation.
4. The method of claim 1 , in which the first format is a three-dimensional bounding box and the second format is a two-dimensional bounding box.
5. The method of claim 1 , in which the ground truth annotation of the real element is annotated by a human annotator.
6. The method of claim 1 , further comprising planning a route for a robotic device via the adjusted machine learning model.
7. An apparatus for training a machine learning model, comprising:
a memory, and
at least one processor coupled to the memory, the at least one processor configured:
to receive real data comprising a real element in a real environment;
to annotate the real element with a first annotation based on predicted attributes of the real element, the first annotation having a first format;
to determine a second format based on an existing format of a ground truth annotation of the real element, the first format and the second format representing different formats of a same annotation type;
to convert the first format of the first annotation to the determined second format; and
to adjust parameters of the machine learning model to minimize a difference between values of the ground truth annotation of the real element and the converted first annotation.
8. The apparatus of claim 7 , in which the at least one processor is further configured:
to receive simulated data comprising a simulated element in a simulated environment; and
to annotate the real element with a second annotation based on predicted attributes of the simulated element, the second annotation having the first format.
9. The apparatus of claim 8 , in which the at least one processor is further configured to adjust the parameters to minimize a difference between values of a ground truth annotation of the simulated element and the second annotation.
10. The apparatus of claim 7 , in which the first format is a three-dimensional bounding box and the second format is a two-dimensional bounding box.
11. The apparatus of claim 7 , in which the ground truth annotation of the real element is annotated by a human annotator.
12. The apparatus of claim 7 , in which the at least one processor is further configured to plan a route for a robotic device via the adjusted machine learning model.
13. A non-transitory computer-readable medium having program code recorded thereon for training a machine learning model, the program code executed by a processor and comprising:
program code to receive real data comprising a real element in a real environment;
program code to annotate the real element with a first annotation based on predicted attributes of the real element, the first annotation having a first format;
program code to determine a second format based on an existing format of a ground truth annotation of the real element, the first format and the second format representing different formats of a same annotation type;
program code to convert the first format of the first annotation to the determined second format; and
program code to adjust parameters of the machine learning model to minimize a difference between values of the ground truth annotation of the real element and the converted first annotation.
14. The non-transitory computer-readable medium of claim 13 , in which the program code further comprises:
program code to receive simulated data comprising a simulated element in a simulated environment; and
program code to annotate the real element with a second annotation based on predicted attributes of the simulated element, the second annotation having the first format.
15. The non-transitory computer-readable medium of claim 14 , in which the program code further comprises program code to adjust the parameters to minimize a difference between values of a ground truth annotation of the simulated element and the second annotation.
16. The non-transitory computer-readable medium of claim 13 , in which the first format is a three-dimensional bounding box and the second format is a two-dimensional bounding box.
17. The non-transitory computer-readable medium of claim 13 , in which the ground truth annotation of the real element is annotated by a human annotator.