How tructrantrung/voguevibe-vto works
Project Overview & Main Components
• This is a full-stack web-based AI platform for fashion recommendations and a touchless virtual try-on feature. It integrates computer vision models (YOLO and DeepFace) for detecting clothing and user attributes, together with deep learning virtual try-on (VITON-HD) modules.
• User Interfaces – Includes the main frontend and the admin panel.
• Backend API – The central Node.js server that handles business logic, authentication (using Firebase), and database interactions (with MongoDB).
• Specialized AI/ML Services:
• Recommendation API: Incorporates YOLO-based clothing item detection (see Web/API_RECOMENDATION).
• Virtual Try-On API: Involves the VITON-HD modules, custom dataloaders, and supporting python utilities (see Web/API_Virtual_Try_On).
• Database & Feature Storage – MongoDB for user and product data along with precomputed features (found in Features_DataBase) that aid in matching similar fashion items.
Analyzing the File Structure for Clues
• The "Web" folder contains all web-focused components:
• /frontend: Contains the customer-facing React-based application.
• /admin: Contains a separate admin panel with its own build configurations and UI components.
• /backend: A Node.js-powered backend server that connects to other services and handles API requests.
• /API_RECOMENDATION and /API_Virtual_Try_On: These directories host Python-based microservices dedicated to AI operations. They include their own supporting libraries, utilities, and pre-trained model files (such as “yolov8n.pt” and the VITON-HD directory).
• There are additional directories like “Features_DataBase” (providing precomputed features and labels for recommendation matching) and “Slide” (for presentation materials).
Establishing Relationships & Data Flow
• User Interaction: A user uploads an image or selfie through the frontend. In parallel, the admin panel provides management functions for products.
• Frontend -> Backend: The frontend components send HTTP requests for login, product recommendation, or virtual try-on processing.
• Backend Service: Acts as the central hub receiving requests, interfacing with the MongoDB database and Firebase for authentication and real-time data sync.
• AI API Microservices:
• Recommendation API (Web/API_RECOMENDATION): Receives image data, triggers the YOLO model to detect clothing items, and then fetches matching products from the database using precomputed feature similarity (using the DINOv2-based features stored under Features_DataBase).
• Virtual Try-On API (Web/API_Virtual_Try_On): Accepts user images, processes them through the VITON-HD pipeline along with supporting dataloaders and network modules, and produces a virtual try-on output.
• Integrations: External services (Firebase for auth, MongoDB for persistent storage) are configured via the backend and appropriate configuration files (as noted in the README).
Including Relevant Technologies and Frameworks
• Frontend: React (indicated by package.json, JSX files, Tailwind CSS config).
• Backend: Node.js, Express-like service (index.js in the backend folder) with integration for Firebase.
• AI & ML Services: Python is used to run the AI modules (YOLO, VITON-HD). Also note supporting libraries such as those handling deep learning and image processing.
• Database: MongoDB is used as the primary data store.
• Supporting Service: Firebase handles user authentication and perhaps real-time database aspects.