Image augmentation apparatus, control method, and non-transitory computer-readable storage medium
An image augmentation apparatus ( 2000 ) acquires an original training dataset ( 40 ). The original training dataset ( 40 ) includes a ground-view image ( 42 ) and an original aerial-view image ( 44 ). The image augmentation apparatus ( 2000 ) performs an image augmentation including a cropping process and a rotation process on the original aerial-view image ( 44 ) to generate an augmented aerial-view image ( 54 ). In the cropping process, a target region with a circle shape or a regular polygon shape is cropped from the original aerial-view image ( 44 ). In the rotation process, the original aerial-view image ( 44 ) is rotated. An angle of the rotation is a multiple of a center angle of the target region when the target region has a regular polygon shape. The image augmentation apparatus ( 2000 ) outputs an augmented training dataset ( 50 ) including the ground-view image ( 52 ) and the augmented aerial-view image ( 54 ).
1 . An image augmentation apparatus comprising:
at least one processor; and
memory storing instructions,
wherein the at least one processor is configured to execute the instructions to:
acquire an original training dataset including a ground-view image and an original aerial-view image;
perform an image augmentation including a cropping process and a rotation process on the original aerial-view image to generate an augmented aerial-view image, the cropping process being a process by which a target region with a regular polygon shape other than a square is cropped from the original aerial-view image, the rotation process being a process by which the original aerial-view image is rotated, an angle of the rotation being an integer multiple of a center angle of the target region; and
output an augmented training dataset including the ground-view image and the augmented aerial-view image.
2 . The image augmentation apparatus according to claim 1 , wherein the image augmentation further includes generating a plurality of the augmented aerial-view images by cropping a plurality of target regions whose shapes are different from each other, and
wherein the outputting of the augmented training dataset further includes outputting a plurality of the augmented training datasets whose augmented aerial-view images are different from each other.
3 . The image augmentation apparatus according to claim 1 , wherein the image augmentation further includes generating a plurality of the augmented aerial-view images whose angles of rotation are different from each other, and
wherein the outputting of the augmented training dataset includes outputting a plurality of the augmented training datasets whose augmented aerial-view images are different from each other.
4 . The image augmentation apparatus according to claim 1 , wherein the image augmentation further includes generating the augmented aerial-view image having a same size as the original aerial-view image by compensating for discarded pixels through an image inpainting after the cropping process and the rotation process.
5 . The image augmentation apparatus according to claim 1 , wherein the image augmentation further includes performing a circular-shift on the original ground-view image to generate an augmented ground-view image, and
wherein the augmented training dataset includes the augmented ground-view image and the augmented aerial-view image.
6 . A control method performed by a computer, comprising:
acquiring an original training dataset including a ground-view image and an original aerial-view image;
performing an image augmentation including a cropping process and a rotation process on the original aerial-view image to generate an augmented aerial-view image, the cropping process being a process by which a target region with a regular polygon shape other than a square is cropped from the original aerial-view image, the rotation process being a process by which the original aerial-view image is rotated, an angle of the rotation being an integer multiple of a center angle of the target region; and
outputting an augmented training dataset including the ground-view image and the augmented aerial-view image.
7 . The control method according to claim 6 , wherein the image augmentation further includes generating a plurality of the augmented aerial-view images by cropping a plurality of target regions whose shapes are different from each other, and
wherein the outputting of the augmented training dataset further includes outputting a plurality of the augmented training datasets whose augmented aerial-view images are different from each other.
8 . The control method according to claim 6 , wherein the image augmentation further includes generating a plurality of the augmented aerial-view images whose angles of rotation are different from each other, and
wherein the outputting of the augmented training dataset includes outputting a plurality of the augmented training datasets whose augmented aerial-view images are different from each other.
9 . The control method according to claim 6 , wherein the image augmentation further includes generating the augmented aerial-view image having a same size as the original aerial-view image by compensating for discarded pixels through an image inpainting after the cropping process and the rotation process.
10 . The control method according to claim 6 , wherein the image augmentation further includes performing a circular-shift on the original ground-view image to generate an augmented ground-view image, and
wherein the augmented training dataset includes the augmented ground-view image and the augmented aerial-view image.
11 . A non-transitory computer-readable storage medium storing a program that causes a computer to execute to:
acquire an original training dataset including a ground-view image and an original aerial-view image;
perform an image augmentation including a cropping process and a rotation process on the original aerial-view image to generate an augmented aerial-view image, the cropping process being a process by which a target region with a regular polygon shape other than a square is cropped from the original aerial-view image, the rotation process being a process by which the original aerial-view image is rotated, an angle of the rotation being an integer multiple of a center angle of the target region; and
output an augmented training dataset including the ground-view image and the augmented aerial-view image.
12 . The storage medium according to claim 11 , wherein the image augmentation further includes generating a plurality of the augmented aerial-view images by cropping a plurality of target regions whose shapes are different from each other, and
wherein the outputting of the augmented training dataset further includes outputting a plurality of the augmented training datasets whose augmented aerial-view images are different from each other.
13 . The storage medium according to claim 11 , wherein the image augmentation further includes generating a plurality of the augmented aerial-view images whose angles of rotation are different from each other, and
wherein the outputting of the augmented training dataset includes outputting a plurality of the augmented training datasets whose augmented aerial-view images are different from each other.
14 . The storage medium according to claim 11 , wherein the image augmentation further includes generating the augmented aerial-view image having a same size as the original aerial-view image by compensating for discarded pixels through an image inpainting after the cropping process and the rotation process.
15 . The storage medium according to claim 11 , wherein the image augmentation further includes performing a circular-shift on the original ground-view image to generate an augmented ground-view image, and
wherein the augmented training dataset includes the augmented ground-view image and the augmented aerial-view image.