Spaces:
Running
Running
File size: 2,647 Bytes
8ab230d 4ed47f2 891e9de 8ab230d c951530 4ed47f2 8ab230d f069dfc 891e9de 059df5c f069dfc 20a1910 8ab230d 891e9de 20a1910 891e9de f069dfc 891e9de f069dfc 20a1910 f069dfc 20a1910 f069dfc 059df5c 20a1910 f069dfc 20a1910 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 | ---
title: LandmarkDiff
emoji: "\U0001F52C"
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 5.50.0
python_version: "3.11"
app_file: app.py
pinned: true
license: mit
short_description: Facial surgery prediction -- 2D foundation toward 3D
tags:
- medical-imaging
- face
- landmarks
- thin-plate-spline
- surgery-simulation
- facial-analysis
- symmetry
---
# LandmarkDiff
Anatomically-conditioned facial surgery outcome prediction from standard clinical photography.
Upload a face photo, select a surgical procedure, adjust intensity, and see the predicted outcome in real time using thin-plate spline warping on CPU.
## Features
- **Single Procedure**: Predict the outcome of one procedure at a chosen intensity
- **Compare All**: See all six procedures side by side at the same intensity
- **Intensity Sweep**: View a single procedure from 0% to 100% in six steps
- **Symmetry Analysis**: Measure bilateral facial symmetry across five regions, with pre/post comparison
## Supported Procedures
| Procedure | Description |
|-----------|-------------|
| **Rhinoplasty** | Nose reshaping (bridge, tip, alar width) |
| **Blepharoplasty** | Eyelid surgery (lid position, canthal tilt) |
| **Rhytidectomy** | Facelift (midface and jawline tightening) |
| **Orthognathic** | Jaw surgery (maxilla/mandible repositioning) |
| **Brow Lift** | Brow elevation and forehead ptosis reduction |
| **Mentoplasty** | Chin surgery (projection and vertical height) |
## How It Works
1. **MediaPipe landmarks** -- 478-point facial mesh extraction
2. **Anatomical displacement** -- procedure-specific landmark shifts scaled by intensity (0-100)
3. **TPS deformation** -- thin-plate spline warps the image smoothly
4. **Masked compositing** -- blends the surgical region back into the original photo
GPU modes (ControlNet, img2img) with photorealistic rendering are available in the full package.
**Roadmap:** This 2D TPS demo is the foundation. Next up -- 3D face reconstruction from phone video for interactive surgical preview.
## Photo Tips
- Use a front-facing, well-lit photo with a neutral expression
- Remove glasses, hats, or anything covering the face
- Make sure only one face is clearly visible
- At least 256x256 resolution recommended
## Links
- [GitHub](https://github.com/dreamlessx/LandmarkDiff-public)
- [Documentation](https://github.com/dreamlessx/LandmarkDiff-public/tree/main/docs)
- [Wiki](https://github.com/dreamlessx/LandmarkDiff-public/wiki)
- [Discussions](https://github.com/dreamlessx/LandmarkDiff-public/discussions)
**Version:** v0.2.2 | **License:** MIT | For research and educational purposes only
|