Generative AI-based style recommendation using fashion item detection and classification

This research work focuses on the creation of a cutting-edge style recommendation system that uses generative AI and deep learning approaches to analyse fashion photos. The system is intended to process input images, such as selfies or studio-quality photos, and output a text file with extensive feedback on the individual's style and suggestions for improvement. The system consists of two main components: the YOLOv8 convolutional neural network trained on DeepFashion2 dataset, which detects and crops clothing items, and the GPT-4.0 large language model, which generates informative style commentary and recommendations. YOLOv8 is briefly trained on a specific dataset to improve its performance in recognising 10 different types of clothes, while GPT-4.0, which is accessible via the OpenAI API, is charged with giving cohesive and short style suggestions. To evaluate the success of the suggested solution, real experimental trials were conducted at many events in Madrid and Tallinn. Three well-known AI models were used for comparison: OpenAI's GPT-4.0 Vision, Google's Gemini 1.5 Pro as reported by Reid et al. (Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context, 2024), and Anthropic's Claude 3 (Templeton et al. in Tansform Circuits Thread, 2024)-Opus. Participants judged the quality of each model's fashion recommendations. The results showed that GPT-4.0 Vision and Gemini 1.5 Pro had comparable average ratings, indicating higher perceived quality than Claude 3-Opus. This research work demonstrates how cutting-edge computer vision and natural language processing technology may transform personalised fashion advising services, improving accuracy and relevance of style recommendations.