AI-Virtual Wardrobe leverages advanced computer vision and artificial intelligence techniques to revolutionize the virtual clothing try-on experience. The project allows users to virtually try on various clothing items and accessories with high accuracy and realism, transforming the fashion industry's approach to garment customization and personal style.
Modules Implemented
- Clothes Try-On Module
Models Explored: IUV Model, Thin Plate Spline, VGG, Keypoint-Based Warping, SeiveNet, GMM & TOM.
Best Model: ACGPN Model - Achieved the highest realism and fit for virtual try-on.
- Glasses Try-On Module
Model Used: Face Detection: Haar cascade classifier.
Facial Landmark Detection: Haar cascade for eye detection.
Approach: Detects the user's face and eyes, then accurately places the glasses.
- Necklace Try-On Module
Model Used:Facial Landmark Detection: Dlib shape predictor.
Approach:Detects facial landmarks, focusing on the neck region for realistic necklace placement.
- Earring Try-On Module
Model Used:Face and Ear Detection: Haar cascade classifiers for face and ear detection.
Approach:Detects face and ears, then places earrings accordingly for a perfect earring placement.
- Hairwig Try-On Module
Model Used: UNet Architecture: Segments the face into five regions—hair, face, ears, neck, and shirt.
Facial Landmark Detection: Used on the segmented image to generate semantic maps.
Approach: The system overlaps the source hair with the target face, creating a difference mask to fill in the missing sections using the SDEdit method. A smooth transition is achieved by filling in these sections using the fast marching approach.
- Lipstick Try-On Module
Model Used: Lip Landmark Detection: Dlib shape predictor.
Lip Detection: Haar cascade classifier.
Approach: Detects lip region and applies different lipstick colors. The system then recommends various products based on the selected color
Frontend Flask Web App
Home Page of Application
Try On Page of Application
