Cloth-net: improved hybrid adversarial network with dense vision transformer for 2D-3D image classification for accurate cloth recommendation engine

Abstract

Purpose

The aim of this study is to enhance the performance of cloth recommendation systems by proposing a hybrid adversarial network called Cloth-Net, which integrates Dense Vision Transformers for effective 2D-3D image classification.

Design/methodology/approach

Cloth-Net combines the strengths of adversarial networks with Dense Vision Transformers to process both 2D and 3D images for improved classification. The model was trained on a large-scale dataset of clothing images, using a hybrid adversarial approach that enhances both feature extraction and image classification accuracy. The methodology also includes data augmentation and transfer learning techniques to optimize the model’s generalization capability.

Findings

Experimental results demonstrate that Cloth-Net significantly outperforms traditional convolutional neural network-based methods in terms of accuracy, precision, and recommendation quality. The hybrid adversarial framework, together with Dense Vision Transformers, enables the model to better understand complex clothing images, leading to more accurate and personalized recommendations.

Originality/value

This study introduces a novel hybrid adversarial model, Cloth-Net, that uniquely combines Dense Vision Transformers with traditional adversarial networks for the first time in the context of 2D-3D image classification. The findings present a substantial improvement in the performance of cloth recommendation engines, making the proposed model valuable for both academic research and practical applications in fashion technology.

Citation

Mrinali, A. and Gupta, P. (2025), "Cloth-net: improved hybrid adversarial network with dense vision transformer for 2D-3D image classification for accurate cloth recommendation engine", International Journal of Clothing Science and Technology, Vol. ahead-of-print No. ahead-of-print. https://doi.org/10.1108/IJCST-05-2024-0110