FIODC Architecture: The Architecture for Fashion Image Annotation

Abstract

An approach to computer vision known as object detection is utilised to locate items in images or videos. Object detection methods usually use machine learning or deep learning to get effective results. When watching photographs or movies, humans are able to quickly recognise and pin down objects of interest. Object detection attempts to mimic this intelligence through the use of computers. Using this method of localization and identification, object detection is able to count the number of objects in a photo, track their precise locations, label them and monitor their movements. In the case of the fashion sector, this object detection is crucial for identifying fashion clothes and classifying them to enable autonomous labelling. Our innovative multi-stage Fashion Image Object Identification and Classification (FIODC) Architecture for object detection and classification is presented in this study. In order to do the same, we have taken into account the performance of four object detection models on our dataset: SSD Resnet, EfficientDet, CenterNet Resnet and Faster RCNN Resnet. The Faster RCNN Resnet outperforms all other models with the highest MAP and classification accuracy, according to experimental data and comparisons of all four models’ MAPs (Mean Average Precision).