---

name: geometry-rendering description: "Reconstruct 3DMM vertices, serialize mesh outputs, and manage 3DDFA rendering helpers." disable-model-invocation: true metadata: disco-role: operating license: MIT

geometry-rendering

Use this sub-skill for the geometry side of 3DDFA: parameter reconstruction, ROI-to-image conversion, sparse and dense vertices, mesh serialization, pose matrices, depth/PNCC/PAF outputs, and the Cython-backed render path.

Use when

• you have a 62-D parameter vector or cropped ROI and need 68 landmarks or dense vertices;
• you need PLY, OBJ, .mat, depth, PNCC, PAF, or pose artifacts;
• you need to build or troubleshoot utils.cython;
• you need dense render or video assembly helpers.

Start here

• API reference
• Output formats
• Rendering and Cython
• Data artifacts
• Troubleshooting

Bundled helpers

• scripts/smoke_geometry.py
• scripts/images_to_video.py

Route elsewhere

• Full detector/model inference and CLI flags belong in the sibling inference sub-skill.
• Training losses, benchmark scripts, and dataset preparation belong in the training sub-skill.
• ONNX export and the C++ port belong in the C++ sub-skill.

Working facts

• utils.ddfa.reconstruct_vertex is the canonical 3DMM reconstruction entry point.
• The verified smoke for a zero 62-D vector returns sparse (3, 68) and dense (3, 53215) vertices.
• visualize/tri.mat provides the triangle indices used by the renderer paths.
• utils.render and utils.lighting depend on the compiled mesh_core_cython extension.