Object count estimation apparatus, object count estimation method, and computer program product

1. An object count estimation apparatus comprising:

a memory; and

one or more hardware processors coupled to the memory and configured to:

obtain an image;

set, based on the image, a local area size representing a unit of object count estimation in the image; and

estimate an object count in the image for each local area having the set local area size, wherein

the one or more hardware processors carry out the estimation of the object count by using:

a learning model for deriving an estimation result of an object count in an area having the local area size in the image by repeatedly performing a convolution operation on the image,

the image, and

the set local area size; and

the one or more hardware processors:

rebuild a reference learning model into the learning model for deriving the estimation result of a local area having the set local area size, the reference learning model being for deriving, from the image, the estimation result of each of local areas having a plurality of the local area sizes different from each other by repeatedly performing a convolution operation on the image, and

carry out the estimation of the object count in the image by using the post-rebuilding learning model and the image.

2. The apparatus according to claim 1 , wherein the one or more hardware processors:

learn the reference learning model for deriving the estimation result from a teacher image by using teacher data representing correspondence between

the teacher image including position information of objects and distribution information representing existence probability distribution of objects, and

an object count estimated from each reduced image obtained by reducing the teacher image to have one of a plurality of the local area sizes different from each other;

rebuild the reference learning model into the learning model for deriving the estimation result estimated from the reduced images having the set local area size; and

carry out the estimation of the object count in the image by using the post-rebuilding learning model and the image.

3. The apparatus according to claim 1 , wherein the one or more hardware processors:

output an output image including a list of the estimation results obtained by estimating an object count in the image for each local area having one of a plurality of the local area sizes different from each other;

receive a user selection on the estimation result included in the output image; and

carry out the setting of the local area size based on the received estimation result.

4. The apparatus according to claim 1 , wherein the one or more hardware processors carry out the setting of the local area size by setting a local area size that corresponds to an estimation result closest to a correct object count in the image obtained in advance from among the estimation results obtained by estimating an object count in the image for each of local areas having a plurality of the local area sizes different from each other.

5. The apparatus according to claim 1 , wherein the one or more hardware processors:

set a plurality of the local area sizes different from each other; and

estimate, as the estimation result, a weighted sum of object counts of areas having the set local area sizes in the image, the object counts being estimated for each set local area size.

6. The apparatus according to claim 1 , wherein the one or more hardware processors:

set the local area sizes different from each other in mutually-different areas in the image; and

estimate, for each of the mutually-different areas in the image, an object count in each local area having the local area size set for the corresponding area.

7. An object count estimation method implemented by a computer, the method comprising:

obtaining an image;

setting, based on the image, a local area size representing a unit of object count estimation in the image; and

estimating an object count in the image for each local area having the set local area size, wherein

the estimating of the object count is carried out by using:

a learning model for deriving an estimation result of an object count in an area having the local area size in the image by repeatedly performing a convolution operation on the image,

the image, and

the set local area size; and

the method further comprises:

rebuilding a reference learning model into the learning model for deriving the estimation result of a local area having the set local area size, the reference learning model being for deriving, from the image, the estimation result of each of local areas having a plurality of the local area sizes different from each other by repeatedly performing a convolution operation on the image, and

carrying out the estimation of the object count in the image by using the post-rebuilding learning model and the image.

8. A computer program product comprising a non-transitory computer-readable recording medium on which an executable program is recorded, the program instructing a computer to:

obtain an image;

set, based on the image, a local area size representing a unit of object count estimation in the image; and

estimate an object count in the image for each local area having the set local area size, wherein

estimation of the object count is carried out by using:

a learning model for deriving an estimation result of an object count in an area having the local area size in the image by repeatedly performing a convolution operation on the image,

the image, and

the set local area size; and

the program further instructs the computer to:

rebuild a reference learning model into the learning model for deriving the estimation result of a local area having the set local area size, the reference learning model being for deriving, from the image, the estimation result of each of local areas having a plurality of the local area sizes different from each other by repeatedly performing a convolution operation on the image, and

carry out the estimation of the object count the image by using the post-rebuilding learning model and the image.