Learning device, learning method, and storage medium
A learning device includes a memory and a processing circuit. The processing circuit: (a) obtains, from the memory, a first computational imaging image which includes an object, the first computational imaging image including a plurality of first pixels; (b) obtains, from the memory, a captured image which includes the object, the captured image including a plurality of second pixels; (c) obtains an identification result of identifying the object included in the captured image; (d) generates, with reference to correspondences between the plurality of first pixels and the plurality of second pixels, an identification model for identifying the first computational imaging image based on the identification result of identifying the object included in the captured image; and (e) outputs the identification model to an image identification device which identifies a second computational imaging image.
1. A learning device comprising:
a memory; and
a processing circuit, wherein
the processing circuit:
(a) obtains, from the memory, a first computational imaging image which includes a first object and surrounding environment of the first object, the first computational imaging image including a plurality of first pixels, the first computation imaging image being an image that is visually unrecognizable by a person, and the first computation imaging image including parallax information indicating that the first object and the surrounding environment of the first object are superimposed multiple times;
(b) obtains, from the memory, a captured image which includes the first object and the surrounding environment of the first object, the captured image including a plurality of second pixels, and the captured image being an image that is visually recognizable by a person;
(c) obtains an identification result of identifying the first object and the surrounding environment of the first object included in the captured image;
(d) generates, with reference to correspondences between (i) the plurality of first pixels included in the first computational imaging image that is visually unrecognizable by a person and (ii) the plurality of second pixels included in the captured image that is visually recognizable by a person, an identification model for identifying the first computational imaging image based on the identification result of identifying the first object and the surrounding environment of the first object included in the captured image, the identification result includes category information indicating categories to which the first object and the surrounding environment of the first object belong; and
(e) outputs the identification model to an image identification device which identifies, in a second computation imaging image, a position of a second object belonging to the same category as the first object, according to whether, for each of pixels of the second computational imaging image, the second object is present in at least one pixel in the second computational imaging image, the second computation imaging image being an image that is visually unrecognizable by a person, and the second computation imaging image including parallax information indicating that the second object and the surrounding environment of the second object are superimposed multiple times.
2. The learning device according to claim 1 , wherein
the identification result includes positions of the first object and the surrounding environment of the first object in a plane.
3. The learning device according to claim 1 , wherein
the identification result includes positions of the first object and the surrounding environment of the first object in a depth direction.
4. The learning device according to claim 1 , wherein
the first computational imaging image is obtained by capturing an image that includes the first object and the surrounding environment of the first object using a multi-pinhole camera, a coded aperture camera, a light-field camera, or a lensless camera, and
the second computational imaging image is obtained by capturing an image that includes the second object and the surrounding environment of the second object using the multi-pinhole camera, the coded aperture camera, the light-field camera, or the lensless camera.
5. The learning device according to claim 1 , wherein
the captured image is obtained by capturing an image of that includes the first object and the surrounding environment of the first object using a multi-view stereo camera.
6. The learning device according to claim 1 , wherein
an optical axis of a camera used for capturing the first computational imaging image and an optical axis of a camera used for capturing the captured image match.
7. The learning device according to claim 6 , wherein
the optical axis of the camera used for capturing the first computational imaging image and the optical axis of the camera used for capturing the captured image are caused to match by using a beam splitter, a prism, or a half mirror.
8. A learning method, comprising:
(a) obtaining a first computational imaging image which includes a first object and surrounding environment of the first object, the first computational imaging image including a plurality of first pixels, the first computation imaging image being an image that is visually unrecognizable by a person, and the first computation imaging image including parallax information indicating that the first object and the surrounding environment of the first object are superimposed multiple times;
(b) obtaining a captured image which includes the first object and the surrounding environment of the first object, the captured image including a plurality of second pixels;
(c) obtaining an identification result of identifying the object and the surrounding environment of the object included in the captured image, and the captured image being an image that is visually recognizable by a person;
(d) generating, with reference to correspondences between (i) the plurality of first pixels included in the first computational imaging image that is visually unrecognizable by a person and (ii) the plurality of second pixels included in the captured image that is visually recognizable by a person, an identification model for identifying the first computational imaging image based on the identification result of identifying the first object and the surrounding environment of the first object included in the captured image, the identification result includes category information indicating categories to which the first object and the surrounding environment of the first object belong; and
(e) outputting the identification model to an image identification device which identifies, in a second computation imaging image, a position of a second object belonging to the same category as the first object, according to whether, for each of pixels of the second computational imaging image, the second object is present in at least one pixel in the second computational imaging image, the second computation imaging image being an image that is visually unrecognizable by a person, and the second computation imaging image including parallax information indicating that the second object and the surrounding environment of the second object are superimposed multiple times.
9. A non-transitory computer-readable recording medium for use in a computer, the recording medium having a computer program recorded thereon for causing the computer to execute the learning method according to claim 8 .