Flyt is a local, CPU-first, publication-oriented computer vision pipeline and full-stack dashboard designed for tracking and phenotyping Drosophila melanogaster behavior. Built specifically for evolutionary biology and behavioral labs, Flyt provides a transparent, deterministic tracking engine paired with a premium human-in-the-loop validation interface.
Flyt is optimized for tracking two flies (typically male and female mating pairs) inside a square acrylic mating chamber. The system accounts for real-world laboratory assay constraints:
Conceptual sketch of the fruit fly mating chamber and tracking overlays.
- High-Texture Mesh Bottom: Employs morphology closing and adaptive thresholding to model mesh variance without fracturing contours.
- Cotton Plug Filter: Filters out static background objects in the corners (like white cotton plugs) based on strict contour bounding area history.
- Reflective Acrylic Walls: Accommodates reflections by offering configurable Region of Interest (ROI) bounds.
- Size Dimorphism: Track trajectories are maintained across overlaps by leveraging size differences between the larger female and smaller male flies.
- deterministic CV Engine: A lightweight, CPU-friendly OpenCV Python backend. No heavy GPU or deep-learning dependencies are required for baseline tracking.
- Interactive Web Dashboard: React Single Page Application (SPA) styled with custom dark-mode aesthetics, featuring rich Recharts visualizations for velocity, inter-fly proximity, and spatial heatmaps.
- Human-in-the-Loop Validation: Review automatically detected courtship bouts, seek directly to relevant video timestamps, and confirm or reject annotations to generate publication-grade metrics.
- GraphPad Prism & PDF Export: One-click formatting of raw trajectory data into column-oriented CSV formats for GraphPad Prism, and custom print styles to export visual reports directly to PDF.
-
Windows & Linux Parity: Run locally on Windows or offload batch sweeps to Modal serverless cloud infrastructure. PNG-lossless frame pipelines eliminate cross-platform decoding drift, guaranteeing exact
$0$ -diff trajectory parity.
Flyt isolates tracking computation from UI state while ensuring immediate feedback loops. The local setup runs a Node.js Express server to orchestrate processes, write CSV outputs, and spawn the Python OpenCV sub-process:
Whiteboard schematic of Flyt's full-stack architecture and data flow.
git clone https://github.com/sudoax0n/flyt.git
cd flytNavigate to the tracker directory, create a virtual environment, and install dependencies:
cd "source app folder/tracker"
python -m venv venv
# Activate virtual environment
# Windows:
venv\Scripts\activate
# macOS/Linux:
source venv/bin/activate
pip install -r requirements.txtNavigate to the dashboard directory, install Node packages, and launch the concurrently run development servers (Express backend on :3001 and Vite frontend on :5173):
cd "../dashboard"
npm install
npm run devOpen your browser and navigate to http://localhost:5173.
If you are dealing with high-throughput video sweeps and want to scale runs on serverless compute, authenticate your environment and invoke the cloud worker:
# Authenticate with Modal
cd flyt-modal
AUTH-MODAL.bat
# Execute cloud tracking job with lossless PNG parity checks
RUN-MODAL-PARITY.batflyt/
├── assets/ # Sample input videos & sketch-style graphics
├── source app folder/
│ ├── tracker/ # Python OpenCV tracking script & venv
│ │ └── tracker.py # Core CV pipeline (~400 lines)
│ └── dashboard/ # Full-stack React (Vite) + Node (Express) app
│ ├── server.js # Orchestration API & H.264 transcoding job loop
│ └── src/
│ ├── App.jsx # Core React SPA (Visual charts, events reviewer)
│ └── index.css # Premium CSS design system & print styles
├── flyt-modal/ # Serverless tracking configuration for Modal
└── AGENTS.md # Detailed developer handbook & living history
This project is licensed under the MIT License - see the LICENSE file for details.
Flyt was developed in collaboration with researchers at the Evolutionary Biology Lab (Dr. N.G. Prasad's Lab), Indian Institute of Science Education and Research (IISER) Mohali.
For full details on development history, performance benchmarks, and experimental features, please consult:
- AGENTS.md - Project architectural records.
- improvements.md - Active roadmap and backlog status.