Learning method, storage medium, and image processing device
According to one embodiment, a learning method of causing a statistical model for outputting a distance to a subject to learn by using an image including the subject as an input is provided. The method includes acquiring an image for learning including a subject having an already known shape, acquiring a first distance to the subject included in the image for learning, from the image for learning, and causing the statistical model to learn by restraining the first distance with the shape of the subject included in the image for learning.
1. A learning method of causing a statistical model for outputting a distance to a subject to learn by using an image including the subject as an input, the method comprising:
acquiring an image for learning including a subject having an already known shape;
acquiring a first distance to the subject included in the image for learning, from the image for learning; and
causing the statistical model to learn by restraining the first distance with the shape of the subject included in the image for learning, wherein
the causing the statistical model to learn comprises:
correcting the first distance to a second distance, based on the shape of the subject included in the image for learning; and
causing the statistical model to learn the image for learning and the second distance,
the image for learning comprises a first image for learning to which a correct label is assigned and a second image for learning to which the correct label is not assigned,
the first and second images for learning include subjects of the same shape,
the acquiring the first distance comprises acquiring a first distance to the subject included in the first image for learning, from the second image for learning,
the first distance is corrected to a second distance, based on the shape of the subject included in the second image for learning, and
the causing the statistical model to learn comprises updating a parameter of the statistical model so as to minimize a value obtained by adding a difference between a relative value of the second distance and a relative value of the third distance output from the statistical model by inputting the second image for learning to the statistical model, to a difference between a correct label and the third distance output from the statistical model by inputting the first image for learning to the statistical model.
2. The method of claim 1 , wherein
the correcting comprises correcting the first distance to the second distance by fitting a parameter used to express the shape of the subject to the first distance.
3. The method of claim 2 , wherein
the shape of the subject is expressed by any function including the parameter.
4. The method of claim 1 , wherein
the causing the statistical model to learn comprises updating a parameter of the statistical model so as to minimize a difference between the second distance and a third distance output from the statistical model by inputting the image for learning to the statistical model.
5. The method of claim 1 , wherein
the causing the statistical model to learn comprises regularizing the statistical model.
6. The method of claim 5 , wherein
the regularizing the statistical model comprises updating a parameter of the statistical model so as to minimize a difference between a relative value of the second distance and a relative value of a third distance output from the statistical model by inputting the image for learning to the statistical model.
7. The method of claim 1 , wherein
the statistical model is generated by learning bokeh which occurs in an image affected by aberration of an optical system and which changes nonlinearly in accordance with the distance to the subject included in the image.
8. The method of claim 1 , wherein
the statistical model is generated by learning bokeh which occurs in an image generated by light passed through a filter and which changes nonlinearly in accordance with the distance to the subject included in the image.
9. The method of claim 1 , wherein
the acquiring comprises acquiring a distance output from the statistical model by inputting the image for learning to the statistical model.
10. The method of claim 1 , wherein
the acquiring comprises acquiring a distance, based on a marker assigned to the subject included in the image for learning.
11. A non-transitory computer-readable storage medium having stored thereon a computer program which is executable by a computer and causes a statistical model for outputting a distance to a subject to learn by using an image including the subject as an input, the computer program comprising instructions capable of causing the computer to execute functions of:
acquiring an image for learning including a subject having an already known shape;
acquiring a distance to the subject included in the image for learning, from the image for learning; and
causing the statistical model to learn by restraining the acquired distance with the shape of the subject included in the image for learning, wherein
the causing the statistical model to learn comprises:
correcting the first distance to a second distance, based on the shape of the subject included in the image for learning; and
causing the statistical model to learn the image for learning and the second distance,
the image for learning comprises a first image for learning to which a correct label is assigned and a second image for learning to which the correct label is not assigned,
the first and second images for learning include subjects of the same shape,
the acquiring the first distance comprises acquiring a first distance to the subject included in the first image for learning, from the second image for learning,
the first distance is corrected to a second distance, based on the shape of the subject included in the second image for learning, and
the causing the statistical model to learn comprises updating a parameter of the statistical model so as to minimize a value obtained by adding a difference between a relative value of the second distance and a relative value of the third distance output from the statistical model by inputting the second image for learning to the statistical model, to a difference between a correct label and the third distance output from the statistical model by inputting the first image for learning to the statistical model.
12. An image processing device for causing a statistical model for outputting a distance to a subject to learn by using an image including the subject as an input, the device comprising:
a processor configured to:
acquire the image for learning including a subject having an already known shape;
acquire a distance to the subject included in the image for learning, from the image for learning; and
cause the statistical model to learn by restraining the acquired distance with the shape of the subject included in the image for learning, wherein
the processor is configured to:
correct the first distance to a second distance, based on the shape of the subject included in the image for learning; and
cause the statistical model to learn the image for learning and the second distance,
the image for learning comprises a first image for learning to which a correct label is assigned and a second image for learning to which the correct label is not assigned,
the first and second images for learning include subjects of the same shape,
the processor is configured to acquire a first distance to the subject included in the first image for learning, from the second image for learning,
the first distance is corrected to a second distance, based on the shape of the subject included in the second image for learning, and
the processor is configured to update a parameter of the statistical model so as to minimize a value obtained by adding a difference between a relative value of the second distance and a relative value of the third distance output from the statistical model by inputting the second image for learning to the statistical model, to a difference between a correct label and the third distance output from the statistical model by inputting the first image for learning to the statistical model.