Examples showcase

The repository ships two levels of Python examples: lightweight scripts under pyexamples/, and optional showcase workflows under examples/ that pull in larger third-party stacks (Torch, vismatch, Rerun).

pyexamples/ (fewer dependencies)

Organized by topic (see pyexamples/README.md):

Area	Path
Preprocess / ZED SVO	pyexamples/preprocess/ — video/sequence extractors; zed_svo_extract_stereo.py writes rectified left/right + rig_calibration.json (needs pyzed)
Stereo / calibrated rig	pyexamples/stereo/stereo_rig_reconstruction.py — baseline + intrinsics, vismatch (edm), GlobalRigReconstructor / IncrementalRigReconstructor (see Rigs)
Monocular SfM	pyexamples/sfm/ — OpenCV fountain, Kornia deep matching, LoFTR + Aqualoc
Two-view / Sim(3) / IO / MVS	twoview/, transform/, io/, mvs/

Typical extras: pip install -r pyexamples/requirements.txt. ZED extraction needs the Stereolabs SDK Python API.

Use these when you want a quick sanity check without installing Torch-based matchers (OpenCV fountain / IO exporters need only light deps).

examples/ (showcase)

• vismatch + SfM — Strecha scene root (images/ + K.txt, gt_dense_cameras/*.camera), Sim3 alignment, average camera-center error vs GT, optional Rerun, PLY/recon exports. Sources: examples/vismatch_sfm/ (vismatch, optional Rerun).
• Nerfstudio / 3DGS export — use pyexamples/io/nerfstudio_export_reconstruction.py to write transforms.json from a pt.sfm.Reconstruction; train Splatfacto or other 3D Gaussian splatting pipelines in external projects (e.g. Nerfstudio).

Install optional dependencies:

pip install ".[examples]"

See examples/README.md for setup, licenses (vismatch wraps many models), and CUDA notes.

Related manual chapters

• Building — compile / install pyTheia
• Python API overview — imports and stubs
• Matching — FeatureCorrespondence, ImagePairMatch, pairing helpers
• IO — WriteNerfStudio, WriteColmapFiles, PLY