Add FLAME 3D morphable model fitting from MediaPipe landmarks

[dreamlessx]

What

Fit FLAME 3D morphable model parameters from MediaPipe's 478 face landmarks, producing a textured 3D mesh from a single 2D photo.

Why

This is a key step toward the 3D roadmap. FLAME fitting from 2D landmarks enables:

• 3D-space surgical deformations (anatomically grounded, not pixel-level)
• Multi-view rendering of predicted results
• Interactive 3D previews for patient consultations

Technical context

• MediaPipe outputs 478 3D landmarks (normalized coords + depth estimate)
• FLAME has ~5K vertices; need a landmark-to-vertex correspondence mapping
• Existing tools like DECA or EMOCA solve similar problems
• The challenge is making it work from MediaPipe landmarks specifically (not DLIB 68 or FAN 98)

Deliverables

1. landmarkdiff/flame_fitting.py — FLAME parameter estimation from MediaPipe landmarks
2. Landmark-to-FLAME vertex correspondence mapping
3. Basic textured mesh output (OBJ or PLY)
4. Tests with synthetic landmarks

References

• FLAME: flame.is.tue.mpg.de
• DECA: github.com/YadiraF/DECA
• EMOCA: github.com/radekd91/emoca

Difficulty: 🔴 Advanced — requires 3D vision experience