Quantitative motion analysis for C. elegans and microfilaria: from raw video to publication-ready data, on your desktop.
ParaTracker turns microscopy videos of worms into structured behavioral data. It detects each animal, extracts a skeleton-based set of keypoints along its body, and tracks how that posture deforms over time, giving you per-worm, per-region motion metrics you can compare across conditions. It runs as a local desktop app: drop in a video, watch it process, and download annotated video plus analysis-ready CSVs.
Scoring worm motility by hand is slow, subjective, and hard to reproduce. Existing tools often demand command-line fluency, a specific OS, or a cloud upload of unpublished data. ParaTracker is built for wet-lab researchers who want rigorous, repeatable motion quantification locally, with no coding and no data leaving their machine.
- Two tracking pipelines, one click apart. A classical computer-vision pipeline (adaptive threshold → skeletonization → Hungarian tracking) that needs no training data, and a deep-learning pipeline built on a custom-trained YOLOv8-seg model for translucent, overlapping, or low-contrast specimens.
- Skeleton keypoints, not just centroids. Each worm is reduced to an ordered set of keypoints from head to tail, so you capture body deformation, not merely whether the animal moved.
- Region-resolved motion. Separate metrics for head, mid-body, and tail, plus an overall score, per worm and over time.
- Head/tail correction built in. Auto-orientation is good, but you can flip any worm's head/tail assignment in the UI and everything downstream recomputes.
- Cross-condition comparison. Group videos into conditions and export grouped comparison charts and statistics.
- Runs entirely on your machine. No account, no server, no upload. Ships as a macOS
.app/ DMG and a Windows installer (or portable.exefolder).
 Before/after slider: drag to reveal original vs. tracked video. |
 Skeleton keypoints: color-coded head→tail, with per-worm IDs. |
 Motion analysis: per-worm heatmap and timeline for head, mid-body, and tail. |
 Tunable parameters: keypoints, area threshold, max age, persistence. |
 Condition comparison: aggregate motion across experimental groups. |
 Per-video consistency: spot outliers and batch effects at a glance. |
- Neurobiology / pharmacology labs screening drug or genotype effects on worm locomotion.
- Parasitology labs quantifying microfilaria motility for anthelmintic assays.
- Anyone who needs reproducible, per-region motion metrics without writing tracking code or shipping data to the cloud.
| Layer | Technology |
|---|---|
| Backend | Python 3.11, FastAPI, SQLite |
| Frontend | React, Vite, Recharts |
| CV / scientific | OpenCV, scikit-image, SciPy, NumPy |
| Deep learning | PyTorch, Ultralytics YOLOv8-seg |
| Video | FFmpeg (bundled via imageio-ffmpeg, H.264 transcoding) |
Coming soon: short walkthrough videos covering upload & processing, results & comparison, job management, motion analysis, and export.
If you just want to run the packaged app, download the latest release for your OS (macOS DMG or Windows installer) and follow the Running a build someone sent you section.
To run from source, install Python 3.11 and Node.js 18+, then:
# macOS / Linux
git clone https://github.com/vclab/worm-tracker.git
cd worm-tracker
make weights
make run# Windows (PowerShell)
git clone https://github.com/vclab/worm-tracker.git
cd worm-tracker
.\dev.ps1 weights
.\dev.ps1 runOpen http://127.0.0.1:5173. Full per-OS instructions (prerequisites, command reference, manual run) are in docs/install.md.
- Installation & Development: prerequisites, per-OS setup (Windows, macOS, Linux), command reference, manual run.
- Using the App: UI walkthrough, tracking parameters, export options, where your data lives, keypoints NPZ format, CLI usage.
- Building for Distribution: building the macOS DMG and the Windows exe + installer, running a build someone sent you.
- Troubleshooting & Uninstalling: common problems and how to remove the app and its data.
- Aaveg Shangari (LinkedIn)
- Faisal Qureshi
VCLab, Faculty of Science, Ontario Tech University