RABET is a desktop application for the behavioural annotation of video recordings of animals. Researchers play back a video, tag behaviours in real time with keyboard shortcuts, visualise events on an interactive timeline, and export annotated data for downstream statistical analysis — including built-in inter-rater and intra-rater reliability assessment.
RABET ships as a self-contained binary. No system-wide video runtime (VLC, FFmpeg, codec packs) needs to be installed: the application bundles its own frame-accurate decoder, so the executable simply works on a clean Windows / macOS / Linux machine.
The Annotation view — video player, recording controls, action map, and colour-coded timeline.
Video & annotation
- Frame-accurate playback with single-frame stepping and instant seek
- Configurable keyboard → behaviour mapping (JSON-driven; ships with a sensible default schema for rodent aggression studies)
- Real-time tagging with monotonic time stamping
- Interactive behaviour timeline with auto-scrolling playhead and colour-coded events
- Timed recording sessions with pause / resume and rewind handling
Analysis & export
- Multi-file CSV analysis: per-session and per-interval summaries
- Custom latency and total-time metrics (e.g. attack latency, total aggression) defined declaratively in JSON
- On-screen Summary and Intervals tabs with one-click clipboard export
- Raster-plot visualisation across multiple animals
Reliability assessment
- Summary mode: compare two
summary_table.csvfiles via ICC(2,1), Pearson correlation, and mean absolute difference per metric, with scatter plots - Detailed mode: compare two annotation CSVs via time-window-binned Cohen's κ, Krippendorff's α, raw percentage agreement, and a pairwise event-raster overlay
Project management
- Group videos, annotations, and analyses under a single project
- Persistent UI settings (window geometry, last view, recording duration, analysis options) stored in the user app-data directory
Cross-platform
- Self-contained binaries for Windows, macOS (Intel & Apple Silicon), and Linux
RABET is officially distributed via Zenodo, which assigns each release a permanent DOI for citation:
concept DOI — always resolves to the latest version
Open the Zenodo record above and download the archive matching your platform:
| Platform | Asset |
|---|---|
| Windows installer | RABET-Windows-1.3.5-Setup.zip |
| Windows portable | RABET-Windows-1.3.5-portable.zip |
| macOS (Apple Silicon) | RABET-macOS-arm64-1.3.5.dmg |
| macOS (Intel) | RABET-macOS-x86_64-1.3.5.dmg |
| Linux | RABET-Linux-x86_64-1.3.5.AppImage |
The downloads are self-contained — no additional installation step is required. See the User Guide for platform-specific launch instructions.
The macOS builds are unsigned and not notarized. After downloading the DMG,
open it and drag RABET.app to Applications. On first launch, macOS may report
that the app is "damaged" or cannot be verified. This is Gatekeeper quarantine,
not a corrupt download.
The most reliable fix is to open Terminal and run:
xattr -dr com.apple.quarantine /Applications/RABET.appThen open RABET.app normally. You only need to do this once for each
downloaded copy.
A mirror of the same binaries is also published on the GitHub Releases page for convenience.
git clone https://github.com/mi2e-K/RABET.git
cd RABET
# Recommended: pinned conda environment
conda env create -f environment.yml
conda activate rabet_build
# Alternative: pip
# pip install -e .
python main.pyPython: 3.11 or newer is required. The pinned conda environment uses Python 3.12.
🌐 Project website: mi2e-k.github.io/RABET — searchable docs site with the full user guide, dark-mode support, and English / Japanese language switcher.
The markdown source for every page below is also browsable directly on GitHub:
- User Guide (English) — step-by-step walkthrough of every feature
- User Guide (Japanese) — Japanese version user guide
- CSV format specification
- Reliability assessment reference
- Build instructions for macOS / Linux
The docs/ directory contains an index of all documentation.
| Platform | Command |
|---|---|
| Windows | python packaging/build_windows_optimized.py |
| macOS | python packaging/build_macos_optimized.py |
| Linux | python packaging/build_linux_optimized.py |
Build outputs land in dist/RABET/ (one-directory mode) or
dist/RABET.exe / dist/RABET.app / dist/RABET-linux-*.tar.gz
(one-file / single-archive mode). See
packaging/README.md for build-script flags and
docs/BUILD_MACOS_LINUX.md for
platform-specific notes.
Annotation exports use the layout below; full documentation lives in
docs/CSV_FORMAT.md.
Metadata
RABET Version,1.3.5
Test Duration (seconds),300
Event,Onset,Offset
Attack bites,12.4123,12.6480
...
Behavior,Duration,Frequency
Attack bites,2.10,3
...
Onsets and offsets are in seconds with 4-decimal precision. In
interval summaries, Duration is the number of seconds overlapping
each interval, and Frequency counts events whose onset falls inside
that interval. Annotation CSVs include a RABET Version row so
downstream tools can detect the producing application; summary CSVs
are kept table-first for easy import into spreadsheet and statistics
software.
If RABET is useful in your research, please cite it. Machine-readable
metadata lives in CITATION.cff; a human-readable form
is:
Mitsui, K. (2026). RABET — Real-time Animal Behavior Event Tagger (Version 1.3.5) [Computer software]. https://github.com/mi2e-K/RABET doi:10.5281/zenodo.15313025
The DOI above is the concept DOI that always resolves to the latest RABET release on Zenodo, so the citation stays valid as new versions are published. Each individual release additionally receives its own version-specific DOI.
A tool paper describing RABET is in preparation. Once published, please cite the paper in addition to (or instead of) the Zenodo deposit.
Released under the MIT License — see LICENSE for the full text.
Bug reports, feature requests, and pull requests are welcome.
- Issues: please include reproducible steps and the application version (Help → About).
- Pull requests: keep them scoped, include a short description, and reference the related issue when applicable.
RABET is built on PySide6, PyAV (FFmpeg python bindings), numpy, pandas, matplotlib, pingouin, krippendorff, filetype, and PyInstaller. We are grateful to the maintainers and contributors of these projects.