Selecting data for deep learning

1. A system for analyzing a data set including a plurality of images, the system comprising:

at least one processor programmed to:

receive a plurality of images acquired by one or more cameras associated with at least one vehicle;

analyze the plurality of images using an active learning system configured to determine a relative priority ranking among the plurality of images, the active learning system including a trained machine learning model configured to receive the plurality of images as an input and to output, for each of the plurality of images, at least a priority ranking indicator based on a combination of a complexity indicator and a diversity indicator of the image, the relative priority ranking indicating an ordered sequence for the plurality of images; and

select, based on the relative priority ranking among the plurality of images, a subset of the plurality of images for use in training a primary image analysis model, wherein the trained machine learning model is trained based at least in part on feedback from the primary image analysis model during training of the primary image analysis model, and wherein during training of the trained machine learning model and the primary image analysis model, the trained machine learning model converges before the primary image analysis model.

2. The system of claim 1 , wherein the relative priority ranking among the plurality of images includes an ordered list of at least two of the plurality of images.

3. The system of claim 1 , wherein the relative priority ranking among the plurality of images includes an ordered ranking associated with at least two of the plurality of images.

4. The system of claim 1 , wherein selecting the subset of the plurality of images includes selecting a percentage of the plurality of images having a predetermined complexity level.

5. The system of claim 1 , wherein selecting the subset of the plurality of images includes selecting a percentage of the plurality of images having a predetermined diversity level.

6. The system of claim 1 , wherein selecting the subset of the plurality of images includes selecting a predetermined number of the plurality of images having a predetermined complexity level.

7. The system of claim 1 , wherein selecting the subset of the plurality of images includes selecting a predetermined number of the plurality of images having a predetermined diversity level.

8. The system of claim 1 , wherein the at least one processor is further programmed to cause the selected subset of the plurality of images to be transmitted to an external system or an external entity.

9. The system of claim 8 , wherein the external system or the external entity is configured to label one or more objects represented in each of the selected subset of the plurality of images.

10. The system of claim 8 , wherein the external system or the external entity is configured to train an external active learning system using the selected subset of the plurality of images.

11. The system of claim 1 , wherein the plurality of images comprise labeled images and the at least one processor is further programmed to input the selected subset of images into the primary image analysis model to train the primary image analysis model.

12. The system of claim 1 , wherein the complexity indicator is based on analysis of edges of the one or more objects represented in the plurality of images.

13. The system of claim 1 , wherein the diversity indicator is based on analysis of an object type of the one or more objects represented in the plurality of images.

14. The system of claim 1 , wherein the active learning system is further configured to determine the relative priority ranking among the plurality of images based on a determined loss prediction.

15. The system of claim 14 , wherein the determined loss prediction is associated with a predicted error for a classification associated with one or more of the plurality of images.

16. The system of claim 1 , wherein the trained machine learning model includes a neural network.

17. The system of claim 1 , wherein the one or more objects include at least one of a vehicle, a pedestrian, a road surface, or a road edge.

18. The system of claim 1 , wherein the plurality of images are acquired during a plurality of drives by a plurality of vehicles.

19. The system of claim 18 , wherein the plurality of drives occurred during a predetermined period of time.

20. The system of claim 1 , wherein the plurality of images are acquired from a geographical region.

21. The system of claim 20 , wherein the geographical region includes a common road segment.

22. The system of claim 1 , wherein the primary image analysis model is trained to identify objects represented in the subset of the plurality of images.

23. A non-transitory computer readable medium containing instructions that when executed by at least one processor, cause the at least one processor to perform a method for analyzing a data set including a plurality of images, the method comprising:

receiving a plurality of images acquired by one or more cameras associated with at least one vehicle;

analyzing the plurality of images using an active learning system configured to determine a relative priority ranking among the plurality of images, the active learning system including a trained machine learning model configured to receive the plurality of images as an input and to output, for each of the plurality of images, at least a priority ranking indicator based on a combination of a complexity indicator and a diversity indicator of the image, the relative priority ranking indicating an ordered sequence for the plurality of images; and

selecting, based on the relative priority ranking among the plurality of images, a subset of the plurality of images for use in training a primary image analysis model, wherein the trained machine learning model is trained based at least in part on feedback from the primary image analysis model during training of the primary image analysis model, and wherein during training of the trained machine learning model and the primary image analysis model, the trained machine learning model converges before the primary image analysis model.

24. The non-transitory computer readable medium of claim 23 , wherein selecting the subset of the plurality of images includes selecting a percentage of the plurality of images having a predetermined complexity level.

25. The non-transitory computer readable medium of claim 23 , wherein selecting the subset of the plurality of images includes selecting a percentage of the plurality of images having a predetermined diversity level.

26. A method for analyzing a data set including a plurality of images, the method comprising:

receiving a plurality of images acquired by one or more cameras associated with at least one vehicle;

analyzing the plurality of images using an active learning system configured to determine a relative priority ranking among the plurality of images, the active learning system including a trained machine learning model configured to receive the plurality of images as an input and to output, for each of the plurality of images, at least a priority ranking indicator based on a combination of a complexity indicator and a diversity indicator of the image, the relative priority ranking indicating an ordered sequence for the plurality of images; and

selecting, based on the relative priority ranking among the plurality of images, a subset of the plurality of images for use in training a primary image analysis model, wherein the trained machine learning model is trained based at least in part on feedback from the primary image analysis model during training of the primary image analysis model, and wherein during training of the trained machine learning model and the primary image analysis model, the trained machine learning model converges before the primary image analysis model.

27. The method of claim 26 , wherein the plurality of images comprise labeled images and the at least one processor is further programmed to input the selected subset of images into the primary image analysis model to train the primary image analysis model.

28. The method of claim 26 , wherein the active learning system is further configured to determine the relative priority ranking among the plurality of images based on a determined loss prediction.

29. A system for analyzing a data set including a plurality of images, the system comprising:

at least one processor programmed to:

receive a plurality of images acquired by one or more cameras associated with at least one vehicle, the plurality of images including labels associated with one or more objects represented in the plurality of images;

analyze the plurality of images using an active learning system configured to select a subset of the plurality of images, the selected subset of the plurality of images being determined based on a determined loss prediction, wherein the active learning system is trained based at least in part on feedback from a primary image analysis model during training of the primary image analysis model to generate the determined loss prediction for each of the plurality of images; and

output the selected subset of images for use in training the primary image analysis model, wherein during training of the active learning system and the primary image analysis model, the active learning system converges before the primary image analysis model.

30. The system of claim 29 , wherein the at least one processor is further programmed to determine the selected subset of images based on at least one indicator, determined for each of the plurality of images, of a diversity level associated with the representations of one or more objects represented in the plurality of images.

31. The system of claim 29 , wherein the at least one processor is further programmed to determine the selected subset based on a determined relative priority ranking among the plurality of images, the relative priority ranking indicating an ordered sequence for the plurality of images.

32. A non-transitory computer readable medium containing instructions that when executed by at least one processor, cause the at least one processor to perform a method for analyzing a data set including a plurality of images, the method comprising:

receiving a plurality of images acquired by one or more cameras associated with at least one vehicle, the plurality of images including labels associated with one or more objects represented in the plurality of images;

analyzing the plurality of images using an active learning system configured to select a subset of the plurality of images, the selected subset of the plurality of images being determined based on a determined loss prediction, wherein the active learning system is trained based at least in part on feedback from a primary image analysis model during training of the primary image analysis model to generate the determined loss prediction for each of the plurality of images; and

outputting the selected subset of images for use in training the primary image analysis model, wherein during training of the active learning system and the primary image analysis model, the active learning system converges before the primary image analysis model.

33. The non-transitory computer readable medium of claim 32 , wherein the primary image analysis model is trained to identify objects represented in the subset of the plurality of images.

34. A method for analyzing a data set including a plurality of images, the method comprising:

receiving a plurality of images acquired by one or more cameras associated with at least one vehicle, the plurality of images including labels associated with one or more objects represented in the plurality of images;

analyzing the plurality of images using an active learning system configured to select a subset of the plurality of images, the selected subset of the plurality of images being determined based on a determined loss prediction, wherein the active learning system is trained based at least in part on feedback from a primary image analysis model during training of the primary image analysis model to generate the determined loss prediction for each of the plurality of images; and

outputting the selected subset of images for use in training the primary image analysis model, wherein during training of the active learning system and the primary image analysis model, the active learning system converges before the primary image analysis model.

35. The method of claim 34 , wherein the active learning system is further configured to determine a relative priority ranking among the plurality of images based on the determined loss prediction.

36. The method of claim 34 , wherein the determined loss prediction is associated with a predicted error for a classification associated with one or more of the plurality of images.