True positive transplant

1. A computer-implemented method comprising:

locating, by a computing system, a foreground object disposed within a seed image, wherein the computing system includes an initial set of images for training a predictive model;

identifying, by the computing system, an object class corresponding to the foreground object;

based on the identified object class, determining, by the computing system, a target value for an object property of the foreground object;

applying, by the computing system, a transformation function to transform the foreground object into a transformed object, wherein the transformation function modifies the object property of the foreground object from having an initial value to having the target value;

transplanting, by the computing system, the transformed object into a background image so as to produce an augmented image; and

augmenting, by the computing system, the initial set of images with the augmented image so as to produce an augmented set of images for training the predictive model.

2. The computer-implemented method of claim 1 , wherein the object property includes a relative size or a relative rotation of the foreground object.

3. The computer-implemented method of claim 1 , wherein the identified object class includes a set of possible target values for the object property, and wherein determining the target value for the object property comprises selecting the target value from the set of possible target values.

4. The computer-implemented method of claim 3 , wherein the set of possible target values includes a probability distribution for the object property, and wherein selecting the target value from the set of possible target values comprises taking a random sample from the probability distribution.

5. The computer-implemented method of claim 1 , wherein transplanting the transformed object into the background image comprises:

determining a target position value for the foreground object; and

placing the transformed object in the background image in accordance with the target position value.

6. The computer-implemented method of claim 5 , wherein the identified object class includes a set of possible target position values for the foreground object, and wherein determining the target position value for the foreground object comprises selecting the target position value from the set of possible target position values.

7. The computer-implemented method of claim 5 , further comprising:

based on the identified object class, establishing, by the computing system, secondary objects of interest; and

determining, by the computing system, that the background image contains at least one of the secondary objects of interest, wherein placing the transformed object in the background image in accordance with the target position value comprises placing the transformed object to be adjacent to at least one of the secondary objects of interest.

8. The computer-implemented method of claim 1 , further comprising:

after augmenting the initial set of images, determining, by the computing system and for each object class of a plurality of object classes, a frequency at which the object class appears in the augmented set of images; and

based on the frequency, determining, by the computing system, a second seed image.

9. The computer-implemented method of claim 8 , wherein determining the second seed image comprises:

making a determination, for the object class, that the frequency at which the object class appears in the augmented set of images is below a threshold; and

based on the determination, selecting, from the augmented set of images, an image that is associated with the object class to be the second seed image.

10. The computer-implemented method of claim 8 , further comprising:

locating, by the computing system, a second foreground object disposed within the second seed image;

identifying, by the computing system, a second object class corresponding to the second foreground object;

based on the identified second object class, determining, by the computing system, a target value for an object property of the second foreground object;

applying, by the computing system, a transformation function to transform the second foreground object into a second transformed object, wherein the transformation function modifies the object property of the second foreground object from having an initial value to having the target value;

transplanting, by the computing system, the second transformed object into the background image so as to produce a second augmented image; and

augmenting, by the computing system, the augmented set of images with the second augmented image so as to produce a second augmented set of images for training the predictive model.

11. The computer-implemented method of claim 1 , wherein the transformation function is an affine image transformation.

12. The computer-implemented method of claim 1 , wherein both the foreground object and the transformed object are associated with the object class.

13. The computer-implemented method of claim 1 , further comprising:

training, by the computing system, the predictive model to determine a respective object class associated with each image in the augmented set of images.

14. The computer-implemented method of claim 1 , further comprising selecting, from the initial set of images, a candidate image to be the seed image.

15. The computer-implemented method of claim 1 , wherein the computing system is a robotic system that operates in a plurality of environments, and wherein the initial set of images are images previously captured by the robotic system as the robotic system operated in the plurality of environments.

16. The computer-implemented method of claim 1 , wherein identifying the object class corresponding to the foreground object comprises:

generating one or more graphical user interfaces that contain data fields for inputting the object class;

transmitting, to a client device, the one or more graphical user interfaces; and

receiving, from the client device, the object class by way of the data fields.

17. The computer-implemented method of claim 1 , further comprising:

based on the identified object class, determining, by the computing system, a second target value for an object property of the foreground object;

applying, by the computing system, a second transformation function to transform the foreground object into a second transformed object, wherein the second transformation function modifies the object property of the foreground object from having an initial value to having the second target value;

transplanting, by the computing system, the second transformed object into the background image so as to produce a second augmented image; and

augmenting, by the computing system, the augmented set of images with the second augmented image so as to produce a second augmented set of images for training the predictive model.

18. The computer-implemented method of claim 1 , further comprising:

based on the identified object class, determining, by the computing system, a target value for a second object property of the foreground object, wherein the transformation function further modifies the second object property of the foreground object from having an initial value to having the target value.

19. A computing system comprising:

an initial set of images for training a predictive model; and

one or more processors configured to cause the computing system to carry out operations comprising:

locating a foreground object disposed within a seed image;

identifying an object class corresponding to the foreground object;

based on the identified object class, determining a target value for an object property of the foreground object;

applying a transformation function to transform the foreground object into a transformed object, wherein the transformation function modifies the object property of the foreground object from having an initial value to having the target value;

transplanting the transformed object into a background image so as to produce an augmented image; and

augmenting the initial set of images with the augmented image so as to produce an augmented set of images for training the predictive model.

20. An article of manufacture including a non-transitory computer-readable medium, having stored thereon program instructions that, upon execution by one or more processors of a computing system that contains an initial set of images for training a predictive model, cause the computing system to carry out operations comprising

locating a foreground object disposed within a seed image;

identifying an object class corresponding to the foreground object;

based on the identified object class, determining a target value for an object property of the foreground object;

applying a transformation function to transform the foreground object into a transformed object, wherein the transformation function modifies the object property of the foreground object from having an initial value to having the target value;

transplanting the transformed object into a background image so as to produce an augmented image; and

augmenting the initial set of images with the augmented image so as to produce an augmented set of images for training the predictive model.