Information processing apparatus and information processing method

An information processing apparatus to adapt an estimation model used for detecting a target object included in an image so as to detect the target object included in the image capturing a specific place, the apparatus comprising:

• a memory;
• and
• processing circuitry configured to:
• acquire a plurality of input images captured at the specific place;
• detect the target object included in the input images by performing a process different from a target object detecting process using the estimation model;
• extract a plurality of first local images each of which includes the target object from the input images, each first local image being one of the input images captured at the specific place;
• generate a plurality of pieces of positive example data, each piece including a target object-included first local image and correct answer information representing the target object included in the first local image;
• extract a plurality of second local images each of which does not include the target object from the input images, each second local image being one of the input images captured at the specific place;
• generates a plurality of pieces of negative example data each of which includes a second local image not including the target object and includes correct answer information representing no presence of the target object in the second local image;
• perform an adaptation of the estimation model so as to detect the target object included in the image capturing the specific place by using a part of the plurality of pieces of positive example data and a part of the plurality of pieces of negative example data;
• calculate an estimation accuracy of the estimation model on which the adaptation has been performed by using another part of the plurality of pieces of positive example data that is not used for the adaptation;
• determine the adaptation as successful when the calculated estimation accuracy is a predetermined accuracy or more;
• determine the adaptation as unsuccessful when the estimation accuracy is less than the predetermined accuracy;
• and
• output a result of determination of the adaptation as successful or as unsuccessful,
• wherein the processing circuitry determines a state in which quality of the images used for the adaptation is poor when performing of the adaptation is unsuccessful and when the number of the part of the plurality of pieces of positive example data used for the adaptation is more than a first threshold, or less than a second threshold, the second threshold being less than the first threshold.

An information processing method to adapt an estimation model used for detecting a target object included in an image so as to detect the target object included in the image capturing a specific place, the method comprising:

• acquiring, by processing circuitry, a plurality of input images captured at the specific place;
• detecting, by the processing circuitry, the target object included in the input images by performing a process different from a target object detecting process using the estimation model;
• extracting, by the processing circuitry, a plurality of first local images each of which includes the target object from the input images, each first local image being one of the input images captured at the specific place;
• generating, by the processing circuitry, a plurality of pieces of positive example data, each piece including a target object-included first local image and correct answer information representing the target object included in the first local image;
• extracting, by the processing circuitry, a plurality of second local images each of which does not include the target object from the input images, each second local image being one of the input images captured at the specific place;
• generating, by the processing circuitry, a plurality of pieces of negative example data each of which includes a second local image not including the target object and includes correct answer information representing no presence of the target object in the second local image;
• performing, by the processing circuitry, an adaptation of the estimation model so as to detect the target object included in the image capturing the specific place by using a part of the plurality of pieces of positive example data and a part of the plurality of pieces of negative example data;
• calculating, by the processing circuitry, an estimation accuracy of the estimation model on which the adaptation has been performed by using another part of the plurality of pieces of positive example data that is not used for the adaptation;
• determining, by the processing circuitry, the adaptation as successful when the calculated estimation accuracy is a predetermined accuracy or more;
• determining, by the processing circuitry, the adaptation as unsuccessful when the estimation accuracy is less than the predetermined accuracy;
• outputting, by the processing circuitry, a result of determination of the adaptation as successful or as unsuccessful;
• and
• determining, by the processing circuitry, a state in which the number of images used for the adaptation is small when performing of the adaptation is unsuccessful and when the number of the part of the plurality of pieces of positive example data used for the adaptation is a first threshold or less, and a second threshold or more, the second threshold being less than the first threshold.