The Desktop App

Everything the library does is reachable from Rust and Python. The desktop app exists so that you do not have to write code to run a calibration or look at its result: point it at a dataset folder, click Run, and get an interactive view of the reprojection errors, the laser residuals, the parameter uncertainty and the 3D rig geometry. It calls the same solvers as the vision-calibration crate.

Launching

The app is built from a local checkout with bun. It uses bun exclusively (never npm, pnpm or yarn); the Tauri CLI is a development dependency, so no global install is needed.

cd app
bun install            # first-time setup, and after package.json changes
bun run tauri dev      # launches the desktop app
bun run tauri build    # bundles the app (installer or binary)

Always use bun run tauri dev. Plain bun run dev starts only the Vite development server in a browser: the file dialog and every command that reaches the Rust backend (running a calibration, computing dense stereo, sniffing a folder) are absent, and the app shows a banner saying so as soon as you try to open a file. Signed installers are not provided.

Mental Model

Five workspaces, reachable from the left rail (hotkeys ⌘1 to ⌘5): Diagnose, 3D viewer, Epipolar, Depth and Run calibration.
One shared store holds the currently loaded *Export JSON plus derived pose and camera indices. Every workspace except Run reads from it; they all react to the same loaded export, the selected pose and the selected camera.
Three ways to fill the store. The top bar's Open Export… button loads a *Export.json file directly. Running a calibration from the Run workspace hands off to Diagnose automatically on success. With a project open, choosing one of its saved runs loads that run's export.
The export needs diagnostics.image_manifest to render anything visual. An export with residual numbers but no manifest produces an error banner ("This export has no image_manifest field") instead of a blank viewer. Runs launched from the Run workspace fill it in; a hand-written export may not have one.

What Is Where

Workspace	Hotkey	Use it to
Run	`⌘5`	configure and launch a calibration from a dataset folder (Running Calibrations)
Diagnose	`⌘1`	look at per-corner residuals on the source images, per-pose statistics and the parameter uncertainty (Inspecting Results)
3D viewer	`⌘2`	see the calibrated rig: camera frustums, target boards, laser planes
Epipolar	`⌘3`	click a point in one image, see its epipolar line in another
Depth	`⌘4`	rectify a stereo pair and compute disparity, depth and a point cloud

The Run workspace is the entry point for a new dataset; the other four work on any export, whether produced by the app or by session.export() in your own code.

Projects

A project is a folder that keeps a dataset's manifest and the runs calibrated from it, so the work survives closing the app:

my-project/
  calibration-project.json   marks the folder as a project; names the dataset folder
  dataset.json               the dataset manifest
  runs/run-001/              manifest.json, config.json, export.json, diagnostics.json, report.html

The Project menu in the top bar does four things:

New project from this dataset… makes a chosen folder a project for the dataset the Run workspace has loaded, whether from a preset, a sniffed folder or a manifest file. The images stay where they are: the project records the dataset folder, relative to the project when it is inside it and absolute otherwise. A dataset.toml you wrote by hand is read once and never rewritten; the project keeps its own dataset.json.
Open project… loads the project's manifest into the Run workspace and lists its runs in a panel on the left; choosing a run loads its export into the viewers. A folder without calibration-project.json is reported as "not a project folder"; the app never guesses one from a folder's contents.
Save manifest writes the Run workspace's edited manifest back to dataset.json.
Recent reopens one of the last projects you opened.

While a project is open, every successful run is saved as the next runs/run-NNN/ folder: the manifest and configuration it ran with, the export, its diagnostics (views used, run time) and report.html, the run as one page to read or share (The Calibration Report). A saved run is never overwritten. Its export.json points back at the dataset's images, so it also opens on its own with Open Export….

The app reads images only from folders you opened (a project's dataset, an export's images); a frame elsewhere fails to load with an error instead of being read silently.

Relation to the Library

The app's forms are generated from the same JSON Schemas as the Rust configuration and manifest types (cargo xtask emit-schemas), so every field the pipeline accepts is editable without hand-writing JSON. The manifest it edits is the dataset manifest; the results it displays are the exports and uncertainty described in this book.