Inspection apparatus, inspection method, and non-volatile storage medium

1. An inspection apparatus comprising:

an input device;

a display device; and

one or a plurality of processors, wherein

the one or the plurality of processors is programmed to execute a method including:

converting an inspection target image representing an inspection target into a virtual good article image by using a learning model, the learning model being trained so that an image representing a good article is generated based on features of a plurality of targets that are determined as good articles, the learning model determining an article to be a good article only when the article does not include a defect that is not designated as an allowable defect,

generating a difference between the virtual good article image and the inspection target image as a defect candidate image,

displaying the defect candidate image on the display device,

receiving, from the input device, designation of an allowable defect from one or a plurality of defect candidates included in the defect candidate image displayed on the display device, the designated allowable defect being one of the one or a plurality of defect candidates in the defect candidate image that is displayed on the display device and being identified as a difference between the virtual good article image and the inspection target image,

restructuring the learning model with the designated allowable defect to update the defect candidate image, and

displaying, on the display device, the updated defect candidate image in which the designated allowable defect is excluded.

2. The inspection apparatus according to claim 1 ,

the method that the one or the plurality of processors is programmed to execute further including: causing the display device to display the virtual good article image.

3. The inspection apparatus according to claim 1 ,

the method that the one or the plurality of processors is programmed to execute further including executing machine learning using images of the plurality of targets that are determined as good articles in appearance inspection and generating the learning model.

4. A method comprising:

converting an inspection target image representing an inspection target into a virtual good article image by using a learning model, the learning model being trained so that an image representing a good article is generated based on features of a plurality of targets that are determined as good articles, the learning model determining an article to be a good article only when the article does not include a defect that is not designated as an allowable defect;

generating a difference between the virtual good article image and the inspection target image as a defect candidate image;

displaying the defect candidate image on a display device;

receiving, from an input device, designation of an allowable defect from one or a plurality of defect candidates included in the defect candidate image displayed on the display device, the designated allowable defect being one of the one or a plurality of defect candidates in the defect candidate image that is displayed on the display device and being identified as a difference between the virtual good article image and the inspection target image;

restructuring the learning model with the designated allowable defect to update the defect candidate image; and

displaying, on the display device, the updated defect candidate image in which the designated allowable defect is excluded.

5. A non-volatile storage medium storing instructions for causing one or a plurality of processors to execute a method, wherein

the method including:

converting an inspection target image representing an inspection target into a virtual good article image by using a learning model, the learning model being trained so that an image representing a good article is generated based on features of a plurality of targets that are determined as good articles, the learning model determining an article to be a good article only when the article does not include a defect that is not designated as an allowable defect,

generating a difference between the virtual good article image and the inspection target image as a defect candidate image,

displaying the defect candidate image on a display device,

receiving, from an input device, designation of an allowable defect from one or a plurality of defect candidates included in the defect candidate image displayed on the display device, the designated allowable defect being one of the one or a plurality of defect candidates in the defect candidate image that is displayed on the display device and being identified as a difference between the virtual good article image and the inspection target image,

restructuring the learning model with the designated allowable defect to update the defect candidate image, and

displaying, on the display device, the updated defect candidate image in which the designated allowable defect is excluded.