Labeling device and learning device

A labeling device according comprising:

• image-signal acquisition circuitry to acquire an image signal indicating an image captured by a camera;
• image recognition circuitry which has learned by machine learning, to perform image recognition on the captured image;
• and
• learning-data-set generation circuitry to generate, by performing labeling on one or more individual objects included in the captured image on a basis of a result of the image recognition, a learning data set including image data corresponding to each of the objects and label data corresponding to each of the objects,
• wherein the image recognition circuitry includes first feature-amount extraction circuitry to generate a first feature map corresponding to the captured image, second feature-amount extraction circuitry to generate a second feature map corresponding to the captured image, and object recognition circuitry to perform object recognition using the first feature map and the second feature map,
• the first feature map corresponds to a foreground mask or corresponds to a mid-level feature corresponding to objectness, and
• the second feature map corresponds to a high-level feature.

A learning device comprising:

• image-signal acquisition circuitry to acquire an image signal indicating an image captured by a camera;
• image recognition circuitry which has learned by machine learning, to perform image recognition on the captured image;
• learning-data-set generation circuitry to generate, by performing labeling on one or more individual objects included in the captured image on a basis of a result of the image recognition, a learning data set including image data corresponding to each of the objects and label data corresponding to each of the objects;
• learning-database update circuitry to update a learning database, by adding the learning data set generated by the learning-data-set generation circuitry to the learning database;
• and
• learning circuitry to perform relearning or additional learning of the image recognition circuitry, using the learning database,
• wherein the learning-database update circuitry assigns a priority to the learning data set generated by the learning-data-set generation circuitry, on a basis of a variation in a distribution of a plurality of learning data sets included in the learning database.