Enhanced training information generation
Systems, methods, and non-transitory computer readable media configured to generate enhanced training information. Training information may be obtained. The training information may characterize behaviors of moving objects. The training information may be determined based on observations of the behaviors of the moving objects. Behavior information may be obtained. The behavior information may characterize a behavior of a given object. Enhanced training information may be generated by inserting the behavior information into the training information.
1. A system comprising:
one or more processors; and
a memory storing instructions that, when executed by the one or more processors, cause the system to perform:
obtaining training information, the training information characterizing behaviors of moving objects, the training information determined based on observations of the behaviors of the moving objects;
obtaining behavior information, the behavior information characterizing a behavior of a given object; and
generating enhanced training information at least in part by inserting the behavior information into the training information, wherein inserting the behavior information into the training information comprises inserting the behavior information based at least in part on a frequency of occurrence of the behavior of the given object.
2. The system of claim 1 , wherein the behavior of the given object includes a rare behavior of the given object.
3. The system of claim 2 , wherein the rare behavior of the given object is not characterized by the training information.
4. The system of claim 2 , wherein the training information defines a three-dimensional point cloud model of a scene and one or more images of the scene.
5. The system of claim 4 , wherein the behavior information characterizing the behavior of the given object includes the behavior information defining a three-dimensional point cloud model of motions of the given object.
6. The system of claim 4 , wherein the given object includes a person, an animal, or a vehicle.
7. The system of claim 1 , wherein the enhanced training information is used to train a machine learning model.
8. The system of claim 7 , wherein the machine learning model is used to control motions of a vehicle.
9. The system of claim 1 , wherein the behavior of the given object comprises an observed behavior of a first portion of the given object and a simulated behavior of a second portion of the given object, and wherein obtaining the behavior information comprises:
obtaining a first portion of the behavior information corresponding to the observed behavior of the first portion of the given object;
generating a second portion of the behavior information at least in part by simulating the simulated behavior of the second portion of the given object; and
modifying the first portion of the behavior information with the second portion of the behavior information.
10. A method implemented by a computing system including one or more processors and storage media storing machine-readable instructions, wherein the method is performed using the one or more processors, the method comprising:
obtaining training information, the training information characterizing behaviors of moving objects, the training information determined based on observations of the behaviors of the moving objects;
obtaining behavior information, the behavior information characterizing a behavior of a given object; and
generating enhanced training information by inserting the behavior information into the training information, wherein inserting the behavior information into the training information comprises inserting the behavior information based at least in part on a frequency of occurrence of the behavior of the given object.
11. The method of claim 10 , wherein the behavior of the given object includes a rare behavior of the given object.
12. The method of claim 11 , wherein the rare behavior of the given object is not characterized by the training information.
13. The method of claim 11 , wherein the training information defines a three-dimensional point cloud model of a scene and one or more images of the scene.
14. The method of claim 13 , wherein the behavior information characterizing the behavior of the given object includes the behavior information defining a three-dimensional point cloud model of motions of the given object.
15. The method of claim 13 , wherein the given object includes a person, an animal, or a vehicle.
16. The method of claim 10 , wherein the enhanced training information is used to train a machine learning model.
17. The method of claim 16 , wherein the machine learning model is used to control motions of a vehicle.
18. The method of claim 10 , wherein the behavior of the given object comprises an observed behavior of a first portion of the given object and a simulated behavior of a second portion of the given object, and wherein obtaining the behavior information comprises:
obtaining a first portion of the behavior information corresponding to the observed behavior of the first portion of the given object;
generating a second portion of the behavior information at least in part by simulating the simulated behavior of the second portion of the given object; and
modifying the first portion of the behavior information with the second portion of the behavior information.
19. A non-transitory computer readable medium comprising instructions that, when executed, cause one or more processors to perform:
obtaining training information, the training information characterizing behaviors of moving objects, the training information determined based on observations of the behaviors of the moving objects;
obtaining behavior information, the behavior information characterizing a behavior of a given object; and
generating enhanced training information by inserting the behavior information into the training information, wherein inserting the behavior information into the training information comprises inserting the behavior information based at least in part on a frequency of occurrence of the behavior of the given object.
20. The non-transitory computer readable medium of claim 19 , wherein the behavior of the given object includes a rare behavior of the given object, the rare behavior of the given object not characterized by the training information.