| CPC G06V 10/764 (2022.01) [G06V 10/255 (2022.01)] | 18 Claims |

|
1. A detection result analysis device comprising processing circuitry
to focus on, as a target layer, each of a plurality of individual layers in an object detection model which detects a target object included in image data and which is constituted using a neural network, and to calculate an evaluation value for each individual target layer, on a basis of a heat map representing an activeness degree per pixel in the image data obtained from an output result of the target layer, and on a detection region where the target object is detected, from a ratio of an activeness degree on an inside of the detection region and an activeness degree on an outside of the detection region, the activeness degree being represented by the heat map, and
to select at least some layers out of the plurality of layers on a basis of the calculated evaluation value for each individual target layer,
wherein the processing circuitry calculates the evaluation value from a proportion of a sum of activeness degrees on the inside of the detection region to a sum of activeness degrees on the outside of the detection region,
wherein the processing circuitry, if the activeness degree is higher than an activeness threshold value, converts the activeness degree into a conversion activeness degree corresponding to the activeness threshold value, and if the activeness degree is equal to or less than the activeness threshold value, converts the activeness degree into a conversion activeness degree corresponding to an activeness threshold value that is one-level lower than the activeness threshold value; and then calculates the evaluation value.
|