Image data generating apparatus generating image data for inspecting external appearance of product

A non-transitory computer readable storage medium storing a set of program instructions installed on and executed by a computer included in an image generating apparatus, the set of first program instructions, when executed by the computer, causing the image generating apparatus to perform:

• acquiring target output image data that is converted from target input image data by a pre-trained model, the target input image data representing a captured image of a target product, the target output image data representing an image of a target product free from defects, the pre-trained model being configured to perform data compression and data decompression on image data inputted to the pre-trained model, wherein the pre-trained model is trained so that when image data representing an input image of the target product is inputted to the pre-trained model, the pre-trained model outputs image data representing an image of the target product free from defects irrespective of whether the input image includes a defect or not;
• generating first data by performing a first generation process on the target input image data, the first generation process including a luminosity adjustment process, the luminosity adjustment process adjusting luminosity in image data inputted to the luminosity adjustment process;
• generating second data by performing a second generation process on the target output image data converted from the target input image data by the pre-trained model in the acquiring, the second generation process including the luminosity adjustment process;
• and
• generating inspection image data by using a difference between the first data and the second data, the inspection image data being used for inspecting an external appearance of the target product.

A method for controlling an image generating apparatus, the method comprising:

• acquiring target output image data that is converted from target input image data by a pre-trained model, the target input image data representing a captured image of a target product, the target output image data representing an image of a target product free from defects, the pre-trained model being configured to perform data compression and data decompression on image data inputted to the pre-trained model, wherein the pre-trained model is trained so that when image data representing an input image of the target product is inputted to the pre-trained model, the pre-trained model outputs image data representing an image of the target product free from defects irrespective of whether the input image includes a defect or not;
• generating first data by performing a first generation process on the target input image data, the first generation process including a luminosity adjustment process, the luminosity adjustment process adjusting luminosity in image data inputted to the luminosity adjustment process;
• generating second data by performing a second generation process on the target output image data converted from the target input image data by the pre-trained model in the acquiring, the second generation process including the luminosity adjustment process;
• and
• generating inspection image data by using a difference between the first data and the second data, the inspection image data being used for inspecting an external appearance of the target product.