Continuous 6-DoF UAV Trajectory Estimation Anchored in Public Orthophotos.

Pick a thumbnail under the video or upload your own data, tune params, then run.

Project Paper Code

Examples — click a thumbnail to load inputs, then Run
Quality vs speed. Matcher settings trade accuracy for runtime on ZeroGPU: fine=precise (best quality — recommended fine matcher (slowest)); coarse=base (balanced — recommended coarse matcher). We recommend precise + base (defaults). Orthophoto resolution (GSD) and DOP/DSM coverage often matter more than matcher speed — low-res or misaligned maps cause large pose errors. Bundled demos ship 0.2 m/px DOP/DSM; max image dim 1280 px.
Upload georeferenced DOP + DSM or pick a bundled example to check resolution.
Fine matcher (RoMa)
Coarse matcher (RoMa)
1 10
1 30
Note: Increasing skip frames drops optical-flow tracking quality and therefore weakens overall OrthoTrack performance. Prefer skip=1 (or low skip) for best results; shorten max duration instead of large skip if you need a faster demo.

Same as CLI --force-calibration. Auto-on when no usable intrinsics.json / meta FoV. Bundled examples ship K/FoV (force_cal off). Cleared when you enter camera params below.

If you know your camera, enter vertical FoV and/or fx fy cx cy (+ image size). This writes a temporary intrinsics.json and disables force-calibration. Leave empty to keep example / auto behavior.

Ready — pick an example or upload data, then run.

Results

Waiting for a run…

Map trajectory

Altitude

3D trajectory

License: CC BY-NC 4.0 (research / non-commercial).