[P2024-16] Online Virtual Dressing Room with Advanced Try-On and Clothing Retrieval Features
Our team developed a categorized dataset of clothing items: tops, bottoms, and dresses. Each item includes flat images, model try-on visuals, and captioned text descriptions generated using Alibaba's large language model Qwen (通义千问) to support text-based search and semantic recommendation.
We use the CLIP (Contrastive Language–Image Pre-training) model to encode both images and user search queries into the same vector space. Text-based search is implemented by comparing the cosine similarity between the query vector and all clothing image vectors. Similarly, click-based recommendation uses a precomputed similarity matrix to find and recommend visually or semantically similar clothing items.
The backend uses a MySQL relational database to manage users, uploaded images, clothing metadata, closet selections, and try-on combinations. The database schema is defined in `backend/database/ovdr_structure.sql`, with sample data in `ovdr_data_only.sql`. An ERD is available as `ERD.png` in the same directory.