Object detection method, object detection apparatus, and non-transitory computer-readable storage medium storing computer program
An object detection method includes inputting an input image to a learned machine learning model and generating a similarity image from an output of at least one specific layer, and generating a discriminant image to which at least an unknown label is assigned, by comparing a similarity of each pixel in the similarity image to a predetermined threshold value.
1. An object detection method of causing one or more processors to detect an object from an input image, by using a vector neural network type machine learning model having a plurality of vector neuron layers, the vector neuron layers having different channels, kernel sizes and strides, the machine learning model being configured to, when a patch image having a predetermined size smaller than the input image is input to the machine learning model, output a determination value indicating that the patch image belongs to one of a plurality of classes, the object detection method comprising:
generating a similarity image by inputting the input image to the machine learning model, and obtaining a similarity from an output of at least one specific layer among the plurality of vector neuron layers for each pixel of the specific layer, the similarity indicating a degree of being similar to a feature of any class among the plurality of classes; and
generating a discriminant image including at least an unknown label by comparing the similarity of each pixel in the similarity image to a predetermined threshold value, and, when the similarity is less than the predetermined threshold value, assigning the unknown label to the pixel, wherein
the generating of the similarity image includes generating an output image from an output of the machine learning model in response to the input of the input image, the output image in which a known label indicating the class to which the output image belongs among the plurality of classes is assigned to each pixel, the generating of the discriminant image includes setting the unknown label to some pixels of the output image with reference to the discriminant image,
two or more specific layers are provided,
the generating of the similarity image includes obtaining the similarity image for each of the two or more specific layers, and
the generating of the discriminant image includes
obtaining the discriminant image for each of the two or more specific layers, and
when the unknown label is assigned to a predetermined number of corresponding pixels among corresponding pixels of the discriminant image for each of the two or more specific layers for each pixel of the output image, setting the unknown label to the pixel of the output image.
2. The object detection method according to claim 1 , wherein
the generating of the discriminant image includes, when the similarity is equal to or more than the predetermined threshold value, assigning a known label to the pixel, the known label corresponding to the class associated with the similarity.
3. The object detection method according to claim 1 , wherein
the generating of the discriminant image further includes performing resolution conversion of causing a resolution of the discriminant image to be equal to a resolution of the output image.
4. A non-transitory computer-readable storage medium storing a program causing a processor to execute an object detection process of detecting an object from an input image using a vector neural network type machine learning model having a plurality of vector neuron layers, the vector neuron layers having different channels, kernel sizes and strides, the machine learning model being configured to, when a patch image having a predetermined size smaller than the input image is input to the machine learning model, output a determination value indicating that the patch image belongs to one of a plurality of classes, the program causes the processor to execute:
a process of generating a similarity image by inputting the input image to the machine learning model, and obtaining a similarity from an output of at least one specific layer among the plurality of vector neuron layers for each pixel of the specific layer, the similarity indicating a degree of being similar to a feature of any class among the plurality of classes; and
a process of generating a discriminant image including at least an unknown label by comparing the similarity of each pixel in the similarity image to a predetermined threshold value, and, when the similarity is less than the predetermined threshold value, assigning the unknown label to the pixel, wherein
the generating of the similarity image includes generating an output image from an output of the machine learning model in response to the input of the input image, the output image in which a known label indicating the class to which the output image belongs among the plurality of classes is assigned to each pixel,
the generating of the discriminant image includes setting the unknown label to some pixels of the output image with reference to the discriminant image,
two or more specific layers are provided,
the generating of the similarity image includes obtaining the similarity image for each of the two or more specific layers, and
the generating of the discriminant image includes
obtaining the discriminant image for each of the two or more specific layers, and
when the unknown label is assigned to a predetermined number of corresponding pixels among corresponding pixels of the discriminant image for each of the two or more specific layers for each pixel of the output image, setting the unknown label to the pixel of the output image.