Machine learning apparatus, machine learning method, and non-transitory computer-readable storage medium
A machine learning apparatus for extracting a region from an input image, comprises: an inference unit configured to output the region by inference processing for the input image; and an augmentation unit configured to, in learning when learning of the inference unit is performed based on training data, perform data augmentation by increasing the number of input images constituting the training data, wherein the augmentation unit performs the data augmentation such that a region where image information held by the input image is defective is not included.
1. A machine learning apparatus for extracting a region from an input image, comprising:
at least one of (a) one or more processors connected to one or more memories storing a program including instructions executed by the one or more processors and (b) circuitry configured to function as:
an inference unit configured to output the region by inference processing for the input image; and
an augmentation unit configured to, in learning when learning of the inference unit is performed based on training data that is constituted by a set of the input image and ground truth data corresponding to the input image and representing an extraction region, perform data augmentation by increasing the number of input images constituting the training data,
wherein the augmentation unit performs the data augmentation such that a region where image information held by the input image is defective is not included.
2. The machine learning apparatus according to claim 1 , wherein
the augmentation unit performs the data augmentation for the training data by at least one augmentation processing of affine transform processing, extraction processing, and signal amount adjustment processing, and
the augmentation unit performs the same augmentation processing for the input image and the ground truth data.
3. The machine learning apparatus according to claim 2 , wherein
the augmentation unit performs the data augmentation by generating an extracted image that extracts a part of the input image constituting the training data, and
a range to acquire the extracted image is limited such that the region where the image information is defective is not included in the extracted image.
4. The machine learning apparatus according to claim 3 , wherein the augmentation unit sets an extractable region in the input image and limits the range to acquire the extracted image.
5. The machine learning apparatus according to claim 3 , wherein as the signal amount adjustment processing, the augmentation unit performs, for the extracted image, multiplication using an arbitrary coefficient and addition using an arbitrary coefficient.
6. The machine learning apparatus according to claim 2 , wherein
the augmentation unit performs the data augmentation by generating an extracted image that extracts a part of a transformed image obtained by affine transform of the input image constituting the training data, and
a range to acquire the extracted image is limited such that the region where the image information is defective is not included in the extracted image.
7. The machine learning apparatus according to claim 6 , wherein the augmentation unit sets an extractable region in the transformed image and limits the range to acquire the extracted image.
8. The machine learning apparatus according to claim 7 , wherein the augmentation unit sets the extractable region in accordance with a rotation angle of the input image in the affine transform.
9. The machine learning apparatus according to claim 6 , wherein the augmentation unit sets a parameter representing a magnification factor of the input image in accordance with the rotation angle of the input image in the affine transform.
10. The machine learning apparatus according to claim 9 , wherein the augmentation unit sets the rotation angle and the parameter representing the magnification factor of the input image such that a part of the input image is not made defective by the affine transform.
11. The machine learning apparatus according to claim 1 , wherein the ground truth data is a labeling image formed by labeling the extraction region in the input image using an arbitrary value.
12. The machine learning apparatus according to claim 1 , wherein the ground truth data is coordinate data representing the extraction region in the input image by coordinates.
13. The machine learning apparatus according to claim 1 , wherein the ground truth data is data that specifies a boundary of the extraction region in the input image by a line or a curve.
14. The machine learning apparatus according to claim 1 , wherein the inference unit performs the inference processing based on a learned parameter acquired based on the learning.
15. The machine learning apparatus according to claim 1 , wherein the machine learning apparatus extracts the region from the input image based on supervised learning using a convolutional neural network.
16. The machine learning apparatus according to claim 1 , wherein the input image is an image captured using a radiation imaging system, and
the region is an irradiation field irradiated with radiation by the radiation imaging system.
17. The machine learning apparatus according to claim 16 , wherein in a use environment of a user, the inference unit performs the inference processing based on a result of learning to which a set of the image captured using the radiation imaging system and data of the irradiation field corresponding to the image is added as training data, and a result of learning performed in advance.
18. A radiation imaging system comprising:
the machine learning apparatus according to claim 1 ; and
a radiation detection apparatus that detects radiation, wherein the radiation detection apparatus are communicatively connected to the machine learning apparatus.
19. A machine learning method by a machine learning apparatus including inference unit configured to output a region by inference processing for an input image and configured to extract the region from the input image, comprising
performing, in learning when learning of the inference unit is performed based on training data that is constituted by a set of the input image and ground truth data corresponding to the input image and representing an extraction region, data augmentation by increasing the number of input images constituting the training data,
wherein the data augmentation is performed such that a region where image information held by the input image is defective is not included.
20. A non-transitory computer-readable storage medium storing a program for causing a computer to execute the machine learning method according to claim 19 .
21. A machine learning apparatus for extracting a region that is an irradiation field irradiated with radiation by a radiation imaging system from an input image that is an image captured using the radiation imaging system, the machine learning apparatus comprising:
at least one of (a) one or more processors connected to one or more memories storing a program including instructions executed by the one or more processors and (b) circuitry configured to function as:
an inference unit configured to output the region by inference processing for the input image; and
an augmentation unit configured to, in learning when learning of the inference unit is performed based on training data, perform data augmentation by increasing the number of input images constituting the training data,
wherein the augmentation unit performs the data augmentation such that a region where image information held by the input image is defective is not included.
22. A radiation imaging system comprising:
the machine learning apparatus according to claim 21 ; and
a radiation detection apparatus that detects radiation, wherein the radiation detection apparatus are communicatively connected to the machine learning apparatus.
23. A machine learning method by a machine learning apparatus including inference unit configured to output a region that is an irradiation field irradiated with radiation by the radiation imaging system by inference processing for an input image that is an image captured using a radiation imaging system and configured to extract the region from the input image, comprising
performing, in learning when learning of the inference unit is performed based on training data that is constituted by a set of the input image and ground truth data corresponding to the input image and representing an extraction region, data augmentation by increasing the number of input images constituting the training data,
wherein the data augmentation is performed such that a region where image information held by the input image is defective is not included.
24. A non-transitory computer-readable storage medium storing a program for causing a computer to execute the machine learning method according to claim 23 .
25. An information processing apparatus comprising:
at least one of (a) one or more processors connected to one or more memories storing a program including instructions executed by the one or more processors and (b) circuitry configured to function as:
an inference unit configured to perform inference processing on a input image using parameter obtained by learning using a training data that is constituted by input images and ground truth data corresponding to the input image, wherein the input images are obtained by augmentation process by extracting a part of an image obtained by a transform process including rotation process.
26. The information processing apparatus according to claim 25 , wherein the inference unit outputs a region that is an irradiation field irradiated with radiation by performing the inference processing.
27. A radiation imaging system comprising:
the information processing apparatus according to claim 25 ; and
a radiation detection apparatus that detects radiation, wherein the radiation detection apparatus are communicatively connected to the information processing apparatus.
28. An information processing method comprising:
performing inference processing on an input image using parameter obtained by learning using a training data that is constituted by input images and ground truth data corresponding to the input image, wherein the input images are obtained by augmentation process by extracting a part of an image obtained by a transform process including rotation process.
29. A non-transitory computer-readable storage medium storing a program for causing a computer to execute the information processing method according to claim 28 .