Virtual Dressing Room
Try on a garment from a single photo or your webcam — a team project I led in my final year.
- Role
- Team lead, team of 3
- Context
- Final-year project, Bangalore Institute of Technology
- Timeline
- Jun 2022 – Mar 2023
- Python
- OpenCV (DNN module)
- Tkinter
- CP-VTON (pretrained)
- OpenPose (pretrained)
- LIP-JPPNet (pretrained)
TL;DR
Problem
Online shoppers can't see how clothes will look on them.
What we built
A desktop app that warps a garment onto a person's photo with a CP-VTON pipeline — pose estimation, human parsing, geometric matching and a try-on renderer. It runs pretrained models on the CPU through OpenCV's DNN module.
My role
I led a three-person team, one short of the usual four. I split the work to balance the heavier load, and integrated the pipeline.
- 01crop to 192×256
- 0218 pose keypoints
- 0320-class body parsing
- 04clothing-agnostic person
- 05garment warp (GMM)
- 06try-on render (TOM)
The pipeline
- Crop the photo to 192×256.
- Find 18 body keypoints (OpenPose).
- Segment the body and clothing into 20 classes (LIP-JPPNet).
- Build a clothing-agnostic representation of the person.
- Warp the garment to the body (the Geometric Matching Module).
- Render the person wearing it (the Try-On Module).
Webcam mode shows a live pose skeleton before capturing a frame.
Limitations
- This is 2-D image try-on, not 3-D or AR.
- The models are pretrained; we didn't train them.
- The research dataset's images can't be shown publicly.