RodentFlow

Pose tracking for dark or white mice across 26 behavior tests on iPhone, iPad and Mac: no GPU, no labeling, no upload. New in 2.1: group behavior.

Pose tracking and the standard measures of 26 behavioral tests for dark or white (albino) laboratory mice, on the iPhone, iPad or Mac you already have. No GPU workstation, no frame labeling or model training, no coding and no upload: import an overhead video, calibrate the arena, and get the test's measures, figures and group statistics, with English data and methods text ready for R, Python or Prism.

TRACKING • Pose: six keypoints per frame (nose, both ears, neck, body center, tail base) from on-device Core ML models, with separate models for dark and for white (albino) mice; choose the coat color for each video. Nose position and head direction make object exploration, sniffing, head dips and edge peeks measurable. • Two mice with A / B identities for social and aggression tests. One dark and one white mouse, the usual pairing in resident–intruder and social defeat experiments, are told apart by coat color without marking. In the review, swap A / B from any frame or over a chosen range. Two white mice are not supported yet. • Group behavior (preview): 3–10 dark mice in one arena, with contact, clustering, nearest-neighbor distance and approach–avoidance. • Classic centroid tracking (10 Hz), as in RodentFlow 1.x, and a multi-chamber tail suspension test that scores several chambers at once.

TESTS Open field (center and corner zones), elevated plus maze, elevated zero maze, light–dark box, novelty-suppressed feeding, visual cliff, marble burying, novel object recognition, three-chamber sociability, social interaction (CSDS), dyadic social interaction, resident–intruder, group behavior, Y-maze alternation, T-maze with VTE, Barnes maze, Morris water maze, passive avoidance, step-down avoidance, platform-mediated avoidance, place preference (RTPP / CPP), fear conditioning (freezing), looming escape, rotation, head direction and spatial exploration, plus the multi-chamber tail suspension test and manual-entry tests.

BUILT FOR A LAB WORKFLOW • Only the calibrated arena is analyzed; tracking that leaves the arena is flagged in the review and in the data. • Every behavior follows a stated rule with adjustable thresholds (for example nose distance and head direction for object exploration), and the methods text records the values used. Change zones and thresholds after tracking; results update without tracking again. • Review every analysis frame by frame, score behaviors by hand and see how well they agree with the automatic scoring. • Heat maps on the original frame, trajectories, timelines and time courses; switch off any figure element before export. • Cover fixed areas of a video with patches and save a copy.

GROUP STATISTICS • The test is chosen from the design and the type of measure, with n = animals: assumption checks and effect sizes for continuous measures, proportions with Fisher's exact or chi-square tests and Wilson intervals for yes/no measures, Kaplan–Meier and log-rank for censored latencies. The methods text states what was done. • Publication figures (PDF / PNG) with captions.

EXPORTS Per-frame data, behavior segments (BORIS-style event table), DeepLabCut-format keypoints, statistics and methods text, always in English whatever the interface language (English / 简体中文).

IPHONE, IPAD AND MAC One purchase covers all devices. Analyses made on the phone appear on the Mac through a sync folder you choose, for example in iCloud Drive. The Mac version handles long recordings and batch analysis, and recomputes imported analyses with the same code to check that the numbers agree.

PRIVACY No account, no data collection. Sync files go only to the folder you choose.

RESEARCH USE Behavior measures are algorithmic estimates. Validate each test, camera setup and strain against manual scoring before drawing conclusions. When animals overlap, two-mouse and group identities can switch; the review marks such moments for checking. Group dominance rankings are a preview and have not been validated against manual scoring.