2ndICASE23B-Out-of-Stock Empty Space Detection Using YOLOv7 for Retail Products on Shelf for Inventory Management

The significant flow of consumers returning to the retail store after the COVID-19 disaster may reflect a new start. To convert visitors into paying customers, it is essential to have a high level of shelf availability in addition to significant store traffic. Using pen and paper to keep track of inventory is more than just time-consuming. It can also result in errors such as stock-outs. This will result in a decrease in sales and consumer happiness, hence lowering customer loyalty. This work focuses on detecting out-of-stock (OOS) products on the shelf of a retail store, investigating the viability of using object recognition techniques to identify empty or almost empty spots on the shelf. This computer vision employs a single-stage framework for object detection based on the YOLOv7 architecture. The model was trained using the SKU-110k dataset. In the experimental study, the effectiveness of the proposed YOLOv7 method was verified on image datasets with the medium size YOLOv7 model. This model will be trained using fixed-resolution images and multi-resolution images. The experimental result shows that the model YOLOv7 training with multi-resolution ideas outperforms the training using fixed-resolution images in terms of precision by 2.81%. This experiment proved that some fine-tuning to the model increased the OOS space detection performance.