Abstract: A training device acquires a degree of similarity between a correct answer region indicating a region of an object in an image and each of a plurality of anchor boxes set in advance in an image, selects, among the plurality of anchor boxes, an anchor box for which the degree of similarity is greater than or equal to a predetermined threshold, for the correct answer region, and performs training of a neural network model for detecting the object, based on the correct answer region and the selected anchor box. If a maximum number of anchor boxes have been selected for the correct answer region, the training device changes the maximum number of the anchor boxes for the correct answer region based on the degrees of similarity acquired for the anchor boxes selected for the correct answer region.

Abstract: From an input image, a candidate area in which an object exists is acquired. Furthermore, a confidence score indicating a possibility that the object exists and a class probability are acquired. An overlap ratio between candidate areas is calculated for each combination of two arbitrary detection results, and a combination of detection results with an overlap ratio equal to or larger than a threshold value is output as an overlap-detected group. A candidate area used as a representative area is determined from detection results included in the overlap-detected group based on a confidence score. Each class index is calculated based on the overlap ratio with respect to the representative area, and a class with the highest class index is determined as the class of the representative area.