Calibration Settings
This article lists every option in Flowbit Calibration's ⚙ Settings panel, and what changing it does to your result.
Open ⚙ Settings from the header.
What each setting does
Setting | What it controls |
|---|---|
Language | English or 中文, applied immediately |
Calibration | GPU on or off, BA backend, matcher, robust features, phone camera model, and Metric scale from, described below |
Reconstruction engine | COLMAP status and download progress, with ↻ Download COLMAP when it is missing or a download failed |
Refinement defaults | Persistent defaults for Refine, including scene grouping. Tick show all for the full list |
Connections | Flowbit Capture host and port, plus a read-only line naming the backend that answered there |
Maintenance | Clear compute cache, clear intermediates, delete this session. Clearing the cache keeps your sessions |
About | The app version |
Sessions
Each attempt is its own session, with its own photos, frames and results. Switch or start one from the header.
Maintenance acts on sessions: it can delete the one you are in, and clearing the compute cache leaves every session intact.
Metric scale from
Calibration has to get its sense of size from somewhere. This setting picks where.
WARNING
Changing this changes the size of every rig you calibrate afterwards. Re-calibrate a rig as a whole rather than one camera at a time.
It also disables Refine on any session already calibrated under the old setting, because refinement has no absolute scale reference of its own. Select Calibrate again and Refine comes back.
auto
The default. It picks the best ruler this particular session has: the board when enough phone photos show it and they agree about its size, otherwise the pose-invariant depth fit, otherwise legacy. The log says which it chose and why, and so does the result. Leave it here unless you have a reason not to.
legacy
Measures scale from the cameras' own depth. That reading is thrown off by each camera's pose error along its viewing direction, and because every camera is affected the same way, averaging them does not remove it.
board
Measures scale from the printed board in your phone photos instead, which that error cannot reach. No camera has to see the board. Phone photos are enough. If the board is missing, or too few photos show it, the run falls back to legacy and says so.
depth-similarity
Also uses the cameras' depth, but in a way that separates the pose error out instead of absorbing it into the answer. Use it when there is no board. A camera whose view is too flat to tell scale apart from a shift is reported as such rather than guessed at.
Checking the scale against your print
When the board was detected, the log prints what the scale in use implies your printed squares measure. If it says 46 mm and you printed 45 mm, the scale is 2% out, so check the print with a tape measure before trusting either number. With no board in the photos there is nothing to compare against, so that line does not appear.
Related
Coordinate systems, for how the anchor you choose sets the origin and the axes.
Calibrate, Refine, and Sync, for where these settings take effect.