Method for improving semantic segmentation accuracy at boundary regions
Disclosed is a method for performing object segmentation using a neural network model, the method performed by one or more processors of a computing device according to an exemplary embodiment of the present disclosure. the method may include: obtaining a training image containing an object; training a neural network model to perform object segmentation using the training image; identifying a boundary region of the object included in the training image; obtaining an additional training image based on the training image and the identified boundary region; and additionally training the trained neural network model based on the additional training image.
1 . A method for performing object segmentation using a neural network model, the method performed by one or more processors of a computing device, the method comprising:
obtaining a training image containing an object;
obtaining a predefined learning rate for a neural network model to perform object segmentation;
training the neural network model to perform object segmentation using the training image based on the predefined learning rate;
finishing training when a learning rate of the neural network model reaches the predefined learning rate;
identifying a boundary region of the object included in the training image;
obtaining an additional training image based on the training image and the identified boundary region; and
additionally training the trained neural network model based on the additional training image,
wherein the identifying the boundary region of the object included in the training image includes:
identifying the boundary region of the object included in the training image based on a shortest distance from a center of the object to an inner boundary of the object and a longest distance from the center of the object to a contour of the object.
2 . The method of claim 1 , wherein the training the neural network model for object segmentation using the training image includes:
training the neural network model for object segmentation using the training image based on at least one of a cross-entropy loss or a dice coefficient loss.
3 . The method of claim 1 , wherein the identifying the boundary region of the object included in the training image includes:
identifying the boundary region of the object included in the training image based on a predefined size with reference to the contour of the object.
4 . The method of claim 1 , wherein the obtaining the additional training image based on the training image and the identified boundary region includes:
obtaining the additional training image by setting a separate weight to the identified boundary region of the training image.
5 . The method of claim 4 , wherein the obtaining the additional training image by setting the separate weight to the identified boundary region of the training image includes:
setting a second weight for the identified boundary region of the additional training image higher than a first weight for an internal region of the object included in the additional training image.
6 . The method of claim 4 , wherein the additionally training the trained neural network model based on the additional training image includes:
additionally training the trained neural network model based on the additional training image with the set weight.
7 . The method of claim 1 , wherein the additionally training the trained neural network model based on the additional training image includes:
additionally training the trained neural network model based on the additional training image using a boundary IOU(Intersection over Union).
8 . A computer program stored in a non-transitory computer-readable storage medium, wherein the computer program causes one or more processors to perform operations for performing object segmentation using a neural network model when the computer program is executed by the one or more processors, the operations comprising:
an operation of obtaining a training image containing an object;
an operation of obtaining a predefined learning rate for a neural network model to perform object segmentation;
an operation of training the neural network model to perform object segmentation using the training image based on the predefined learning rate;
an operation of finishing training when a learning rate of the neural network model reaches the predefined learning rate;
an operation of identifying a boundary region of the object included in the training image;
an operation of obtaining an additional training image based on the training image and the identified boundary region; and
an operation of additionally training the trained neural network model based on the additional training image,
wherein the operation of the identifying the boundary region of the object included in the training image includes:
an operation of identifying the boundary region of the object included in the training image based on a shortest distance from a center of the object to an inner boundary of the object and a longest distance from the center of the object to a contour of the object.
9 . The computer program of claim 8 , wherein the operation of the training the neural network model for object segmentation using the training image includes:
an operation of training the neural network model for object segmentation using the training image based on at least one of a cross-entropy loss or a dice coefficient loss.
10 . The computer program of claim 8 , wherein the operation of identifying the boundary region of the object included in the training image includes:
an operation of identifying the boundary region of the object included in the training image based on a predefined size with reference to the contour of the object.
11 . The computer program of claim 8 , wherein the operation of obtaining the additional training image based on the training image and the identified boundary region includes:
an operation of obtaining the additional training image by setting a separate weight to the identified boundary region of the training image.
12 . The computer program of claim 11 , wherein the operation of obtaining the additional training image by setting the separate weight to the identified boundary region of the training image includes:
an operation of setting a second weight for the identified boundary region of the additional training image higher than a first weight for an internal region of the object included in the additional training image.
13 . The computer program of claim 11 , wherein the operation of additionally training the trained neural network model based on the additional training image includes:
an operation of additionally training the trained neural network model based on the additional training image with the set weight.
14 . The computer program of claim 8 , wherein the operation of additionally training the trained neural network model based on the additional training image includes:
an operation of additionally training the trained neural network model based on the additional training image using a boundary IOU(Intersection over Union).
15 . A computing device comprising:
at least one processor; and
a memory,
wherein the at least one processor is configured to:
obtain a training image containing an object;
obtain a predefined learning rate for a neural network model to perform object segmentation;
train the neural network model to perform object segmentation using the training image based on the predefined learning rate;
finish training when a learning rate of the neural network model reaches the predefined learning rate;
identify a boundary region of the object included in the training image;
obtain an additional training image based on the training image and the identified boundary region; and
additionally train the trained neural network model based on the additional training image, and
wherein the identifying the boundary region of the object included in the training image includes:
identifying the boundary region of the object included in the training image based on a shortest distance from a center of the object to an inner boundary of the object and a longest distance from the center of the object to a contour of the object.