Inventory distribution considering Image-based fashion product variety

Abstract

In the fashion industry, brands have updated the sales season frequently with more varied but smaller production quantities. Due to a limited amount of production, some products are not always supplied to all stores. Therefore, company managers have to make decisions to distribute the products. To satisfy a wide range of customer needs, a visual variety should be considered in the initial distribution. We turned product images into numbers using a convolutional autoencoder. The numeric values were utilized to measure the visual variety. The proposed distribution optimization model maximized the variety of products stores receive while considering operational constraints. We applied the model to a base to verify the applicability of the proposed model. The distribution results were consistent with what the company is pursuing.