Release notes

Open source (October 2026)

The source code of RodentFlow 1.3 and this manual are published at github.com/seesaw-earth/RodentFlow under the MIT License. See Open source & building.

The whole app is available in English and Simplified Chinese: screens, assay and metric names, definitions, methods notes, quality flags, statistics wording, figures and the text summary. It follows the iPhone language. Aa › Language / 语言 opens the setting for choosing a different language for RodentFlow alone.
Stored data are language-independent: result files, JSON keys, assay ids, phases and CSV template columns stay English, so files move freely between phones set to different languages.
Imports accept Chinese column names (编号, 组别, 性别, 天, 试次, 阶段, 日期, 备注), and, when the app runs in Chinese, the Chinese metric names shown in the app.
The import preview lists every recognized column.

No more pauses when choosing a video, opening the preview, finishing an analysis or switching tabs. Copying videos, writing result files and loading checkpoints moved off the main thread, and progress updates redraw only what changed.

Larger, adjustable text (Aa: Small to Largest, or same as iPhone).
A step bar (Start → Set up → Preview → Result) and a back/next bar on every workflow screen.
A reminder to update the preview if the setup changed after it was made.
Batch scoring: Next Animal keeps assay, group, phase and day; Use Previous Setup reuses arena, calibration and settings.
▲▼ navigation between results of one assay, and Compare in Groups from any result.
Section chips in the Groups tab and a section menu on the setup screen.

Fixes found with synthetic videos of known ground truth:

Frames are read on the full 0–255 gray scale, matching the background. A mouse hidden in a covered dark box is no longer "detected".
Resting-animal repair of the background removes "ghosts" left by an animal that rested at one spot for much of the video.
Contrast-weighted centroid: median position error fell from about 1 cm to 0.15–0.5 cm, and path length in mazes is far less inflated.
The motion-index pixel threshold was raised from 12 to 20 gray levels, so compression flicker no longer breaks freezing bouts apart.
Default object-interaction and Barnes hole-zone radii raised from 6 to 8 cm (body-center proxies).

Two tabs: Experiment (per-animal results) and Groups (group comparison).
15 assays plus custom tables, from video, manual entry or CSV/JSON import.
Plus-maze calibration; object, platform and target-hole markers; motion index for freezing.
Results with editable zones and thresholds, recomputed without re-tracking.
Statistics chosen from the design, with figures showing every animal and export to PNG and JSON.

Classic-centroid tracking of open field (rectangle or circle), T-maze and Y-maze videos at 10 Hz, with preview, full analysis, recoverable checkpoints and JSON export.