Fashion Generation and Classification for Morocco Traditional Clothes (MOROFASHION)

Fashion datasets have played a crucial role in advancing clothing recognition systems. While several comprehensive datasets have been developed, there is a notable absence of datasets representing traditional Moroccan fashion. We introduce MOROFASHION, a dataset focusing on Moroccan traditional women's fashion to address this gap. MOROFASHION comprises over 3600 images spanning four distinct fashion item classes: Kmiss, Djellaba, Kaftan, and Takchita. In addition to introducing MOROFASHION, we present image generation and classification experiments using this dataset. For image generation, we utilize a latent diffusion model (SDXL) fine-tuned with Low-Rank Adaptation (LoRA) to generate new images of Moroccan traditional fashion. For image classification, we employ Convolutional Neural Networks (CNNs) and transfer learning techniques. Furthermore, we investigate the impact of combining original and generated data on classification tasks to improve its performance.