Shelf-based product arrangement training data generation device using image synthesis, shelf-based product arrangement training data generation system using image synthesis, shelf-based product arrangement training data generation method using image synthesis, and recording medium

A training data generation method comprising:

• acquiring a shelf image constituting one compartment of a shelf on which a product is displayed, and the shelf image represents a shape of the shelf as at least one of an uneven shape and a planar shape;
• acquiring a product image of the product displayed on the shelf;
• generating training data by synthesizing the shelf image and the product image by at least:
• determining an arrangement of the product image according to at least one of the shape of the shelf and a shape of the product;
• and
• at least one of:
• synthesizing, based on determining the arrangement and that the shape of the shelf comprises the uneven shape, first product images, based on the product image, arranged in a row in the shelf image;
• and
• synthesizing, based on determining the arrangement and that the shape of the shelf comprises the planar shape, second product images, based on the product image, randomly arranged in the shelf image;
• and
• training, using the training data, a machine learning model to recognize, from one or more captured images, products, including the product, arranged on any of the shelf and another shelf.

A non-transitory computer-readable storage medium that stores a training data generation program that causes a computer to enable:

• acquiring a shelf image constituting one compartment of a shelf on which a product is displayed, and the shelf image represents a shape of the shelf as at least one of an uneven shape and a planar shape;
• acquiring a product image of the product displayed on the shelf;
• generating training data by synthesizing the shelf image and the product image by at least:
• determining an arrangement of the product image according to at least one of the shape of the shelf and a shape of the product;
• and
• at least one of:
• synthesizing, based on determining the arrangement and that the shape of the shelf comprises the uneven shape, first product images, based on the product image, arranged in a row in the shelf image;
• and
• synthesizing, based on determining the arrangement and that the shape of the shelf comprises the planar shape, second product images, based on the product image, randomly arranged in the shelf image;
• and
• training, using the training data, a machine learning model to recognize, from one or more captured images, products, including the product, arranged on any of the shelf and another shelf.