Storage medium having stored learning program, learning method, and learning apparatus

1. A non-transitory computer-readable storage medium having stored therein a learning program for causing a computer to execute a process, the process comprising:

receiving a first image that includes an object;

generating a first rectangle in the first image, the first rectangle including therein a figure, that is set in advance to have a first inclination and that represents a gripping position of the object, and having a side parallel to a first direction;

inputting the first image to a model, which outputs, from an input image, a rectangle parallel to the first direction and an inclination, to cause the model to output a second rectangle and a second inclination; and

updating the model such that errors of the second rectangle and the second inclination with respect to the first rectangle and the first inclination respectively decrease.

2. The non-transitory computer-readable storage medium according to claim 1 , wherein

in the generating, a rectangle having a minimum area and a side parallel to the first direction is selected as the first rectangle among rectangles each including therein all vertices of the figure that is a rectangle.

3. The non-transitory computer-readable storage medium according to claim 1 , the process further comprising:

inputting a second image to the model that has been updated in the updating, to cause the model to output a third rectangle and a third inclination; and

detecting a gripping position of an object from the third rectangle and the third inclination.

4. The non-transitory computer-readable storage medium according to claim 3 , the process further comprising:

outputting the gripping position detected in the detecting to a gripping control apparatus that controls a robot for gripping an object.

5. The non-transitory computer-readable storage medium according to claim 1 , wherein the model further outputs, from the input image, a confidence for each class indicating display content in the rectangle in the input image, along with the rectangle and the inclination,

in the inputting, the first image is input to the model to cause the model to output a confidence for each class along with the second rectangle and the second inclination, and

in the updating, the model is updated such that an error between a confidence for a first class set in advance for the figure and the confidence for the second rectangle output by the model further decreases.

6. The non-transitory computer-readable storage medium according to claim 5 , wherein

the model is a model obtained by causing a model of Single Shot MultiBox Detector (SSD) to further output an angle, the model of SSD being configured to output, from an input image, any of bounding boxes that are rectangles set in advance and a confidence for each class,

in the inputting, the first image is input to the model to cause the model to output a bounding box as the second rectangle, an angle as the second angle, and a confidence for each class, and

in the updating, the model is updated such that an error between the bounding box output by the model and the first rectangle, an error between the angle output by the model and the first angle, and an error between the confidence for the first class output by the model and a confidence for the first class set in advance for the figure decrease.

7. A learning method performed by a computer, the method comprising:

receiving a first image that includes an object;

generating a first rectangle in the first image, the first rectangle including therein a figure, that is set in advance to have a first inclination and that represents a gripping position of the object, and having a side parallel to a first direction;

inputting the first image to a model, which outputs, from an input image, a rectangle parallel to the first direction and an inclination, to cause the model to output a second rectangle and a second inclination; and

updating the model such that errors of the second rectangle and the second inclination with respect to the first rectangle and the first inclination respectively decrease.

8. The learning method according to claim 7 , the method further comprising:

inputting a second image to the model that has been updated in the updating, to cause the model to output a third rectangle and a third inclination; and

detecting a gripping position of an object from the third rectangle and the third inclination.

9. The learning method according to claim 8 , the method further comprising:

outputting the gripping position detected in the detecting to a gripping control apparatus that controls a robot for gripping an object.

10. A learning apparatus comprising:

a memory, and

a processor coupled to the memory and configured to:

receive a first image that includes an object;

generate a first rectangle in the first image, the first rectangle including therein a figure, that is set in advance to have a first inclination and that represents a gripping position of the object, and having a side parallel to a first direction;

input the first image to a model, which outputs, from an input image, a rectangle parallel to the first direction and an inclination, to cause the model to output a second rectangle and a second inclination; and

update the model such that errors of the second rectangle and the second inclination with respect to the first rectangle and the first inclination respectively decrease.

11. The learning apparatus according to claim 10 , the processor is further configured to:

input a second image to the model that has been updated in the update, to cause the model to output a third rectangle and a third inclination; and

detect a gripping position of an object from the third rectangle and the third inclination.

12. The learning apparatus according to claim 11 , the processor is further configured to:

output the gripping position detected in the detecting to a gripping control apparatus that controls a robot for gripping an object.