iTrace
github.com
- Free plan
- Linux
- Web
- Self-hosted
What iTrace is
iTrace is an open-source webcam eye-tracking analysis toolkit with a scientific analysis core built on NumPy and SciPy. It calculates gaze in degrees of visual angle and identifies fixations, saccades, and microsaccades using I-VT, I-DT, and Engbert–Kliegl methods. Pupil tools support blink detection and interpolation, preprocessing, and eye-crop segmentation measured in pixels or relative to iris size. Affine screen-space calibration reports calibration error and gaze-quality measures. Optional webcam capture uses OpenCV and MediaPipe iris landmarks, while optional interfaces provide a local HTML app using FastAPI and WebSockets and a Streamlit dashboard. The command-line tool can analyze gaze CSV files and export JSON reports; camera workflows can record gaze and pupil CSV data. Default exports omit raw eye video and saved eye-crop images, although live viewing may send a temporary JPEG crop to a browser over a local WebSocket. The project says its checks use synthetic ground truth, not real eye-tracking data or a reference device. Webcam pupil measurement is a relative proxy, not a diameter in millimetres.
Who it suits
iTrace is intended for science and research users working with webcam gaze analysis. The project presents it as a reference implementation and advises comparison with a reference device or public dataset before reporting device-level accuracy.
Good to know, and what to watch
- Uses several methods to detect gaze events.
- Includes pupil blink detection and interpolation.
- Exports derived data as CSV and JSON.
- Released under the MIT license.
- No real-device validation is claimed.
- Webcam pupil measurement is only a relative proxy.
- Default workflow does not save raw eye video.
Plans
Every plan and priceiTrace (MIT-licensed open-source software)
Open source
Getting started
- Open the iTrace GitHub website
- Install the Python package with uv
- Add optional capture, dashboard, figures, or web extras as needed
- Provide gaze CSV files to the CLI or connect a webcam for recording
- Use the local HTML app or Streamlit dashboard when installed
The Club's review
iTrace is a free, MIT-licensed Python toolkit for analysing gaze and pupil data, with optional webcam capture and local browser interfaces. It suits researchers and developers who need inspectable algorithms, configurable analysis, and a self-hosted workflow. Choose it for method development and exploratory studies; look elsewhere when validated device accuracy or turnkey participant management is the priority.
Overview
The NumPy and SciPy core accepts gaze CSV files from the command line, calculates gaze in degrees of visual angle, and produces JSON analysis reports. Webcam capture is an optional layer rather than a requirement, so existing recordings can be analysed without adopting the capture stack. The project targets science and research use, and its MIT licence permits practical reuse and modification.
Its evidence has a clear boundary. iTrace reports algorithmic checks against synthetic ground truth, but no validation against real eye-tracking data or a reference tracker. Treat it as a reference implementation for comparing methods, not proof of device-level accuracy; serious measurement work should compare outputs with a reference device or public dataset.
Key features
Gaze events and inspection
Fixations and saccades can be detected with I-VT and I-DT, while microsaccades use the Engbert–Kliegl method. Visualizations include gaze traces, x/y time series, speed plots, event rasters, scanpaths, direction polar plots, amplitude histograms, and main-sequence scatter. That range is useful for examining both event classifications and continuous movement, although it still leaves interpretation and methodological choices to the user.
Pupil processing
Pupil tooling covers blink detection and interpolation, preprocessing, and eye-crop segmentation. Size is reported in pixels or relative to iris size, which supports within-session analysis but is not a physical measurement: live webcam pupil output does not infer diameter in millimetres. Calibration uses affine screen-space mapping and reports calibration error and gaze-quality measures, giving researchers diagnostics rather than a guarantee of real-world precision.
Capture, interfaces, and exports
Optional capture combines OpenCV with MediaPipe Face Mesh iris landmarks. A local FastAPI and WebSocket HTML app provides browser viewing, while a Streamlit dashboard adds Plotly-based exploration. The live workflow can record gaze and pupil CSV data and export derived gaze, pupil, and capture-record CSVs alongside a JSON report. By default it does not save raw eye video or persisted eye-crop images; live viewing may send a transient JPEG eye crop to the browser over a local WebSocket.
Python installation is documented through uv, with capture, dashboard, figures, and web components available as optional extras. This modular setup keeps the scientific core lighter, but assembling a complete capture-and-dashboard environment requires choosing and installing those components.
Pricing
iTrace (MIT-licensed open-source software): free (MIT license · Python package). There are no paid tiers or seat charges. The trade-off is that support, deployment, calibration decisions, and validation remain the user's responsibility; this is software for teams able to work with Python and self-hosted tooling rather than buyers seeking a managed service.
Platforms
iTrace runs on Linux and can be self-hosted. Its local HTML interface is accessed through a web browser. The webcam input method is a camera, while analysis can also start from gaze CSV files.
Who it's for
Researchers, students, and developers building or comparing gaze-event and pupil-processing methods will get the most from iTrace's inspectable NumPy/SciPy pipeline and multiple detection algorithms. It is also suitable for teams that need local data handling and derived-file exports without default raw-video storage. It is a poor fit for studies that require millimetre pupil measurements, established real-device validation, or a hosted recruitment and session-management system.
Pros and cons
- Pros: Free MIT licensing removes subscription and seat barriers.
- Pros: I-VT, I-DT, and Engbert–Kliegl methods provide meaningful analytical choice.
- Pros: Optional capture and local FastAPI, WebSocket, and Streamlit interfaces extend CSV analysis into a complete local workflow.
- Pros: Derived CSV and JSON exports, with no default raw eye-video or saved eye-crop files, support privacy-conscious pipelines.
- Cons: No real-device or reference-tracker validation is claimed, so accuracy conclusions require external comparison.
- Cons: Webcam pupil size is only a pixel or pupil-to-iris-relative proxy, not millimetres.
- Cons: Optional components and self-hosting put installation and operational work on the user.
Alternatives
Labvanced is the better choice when a freemium, web-based research platform with a free trial, 300 MB storage, 10 recordings, and one active recruiting link is more useful than a self-hosted Python toolkit.
Chiasm suits teams wanting hosted API and web access with pay-per-use Flex billing at €0.25 per minute and no monthly minimum, or a paid Starter tier.
GazeRecorder is an option for non-commercial webcam gaze recording across Android, iOS, web, and Windows.
GazePlotter is a free, GPLv3 open-source tool for readers who prefer broader desktop and mobile platform coverage.
Hawkeye fits iOS and web projects that can work within a free plan of one project and five sessions or move to its $39.99 monthly Starter plan.
OGAMA is free open-source freeware for Windows users.
OpenGazer is another free, open-source Linux and self-hosted webcam gaze tracker.
PyGaze is a free open-source Python eye-tracking toolbox for Linux, macOS, and Windows.
Verdict
iTrace is a strong fit for technically capable researchers who want a free, inspectable foundation for gaze-event and pupil-analysis experiments, especially when data must stay local. Its breadth and modular capture options are compelling, but the absence of real-device validation and the relative-only webcam pupil measure are decisive limitations. Pick a validated or managed alternative when your conclusions depend on calibrated device accuracy or a ready-made study platform.
The verdict
iTrace offers a substantial set of gaze and pupil analysis methods, with optional capture and local interfaces. Its own stated validation limits make reference comparisons necessary before treating results as device-level accuracy.
Questions people ask
What does iTrace cost?
iTrace is free and released under the MIT license.
Which platforms does it support?
Its listed platforms are Linux, self-hosted deployments, and a web browser through the local HTML interface.
Can it analyze existing gaze data?
Yes. The CLI analyzes gaze CSV files and can export JSON reports.
Does it measure pupil diameter in millimetres?
No. Live webcam pupil measurement is a relative algorithmic proxy and does not infer millimetres.
What does the default export contain?
It writes derived gaze, pupil, and capture CSV files plus analysis reports, without raw eye video or persisted eye-crop images by default.
Is webcam capture required?
No. The NumPy/SciPy analysis core can process gaze data, while webcam capture is optional.
What the maker publishes
- Accuracy limit
- The live webcam pupil measurement is a relative proxy and does not infer pupil diameter in millimetres. source
- Accuracy limitation
- The project states that live webcam pupil measurements are a relative algorithmic proxy and do not infer millimetres. source
- Analysis core
- The scientific analysis core uses NumPy and SciPy, with webcam capture provided as an optional layer. source
- Analysis methods
- Its analysis includes I-VT and I-DT event detection, Engbert–Kliegl microsaccade detection, and pupil blink detection and interpolation. source
- Calibration
- It supports affine screen-space calibration and reports calibration error and gaze-quality measures. source
- Capture and interfaces
- Optional components use OpenCV and MediaPipe for capture, FastAPI and WebSockets for a local HTML interface, and Streamlit with Plotly for a dashboard. source
- Capture integrations
- Optional webcam capture uses OpenCV and MediaPipe Face Mesh iris landmarks. source
- Data handling
- The README says the default export writes derived gaze, pupil, and capture CSVs without writing raw eye video or persisted eye-crop images. source
- Exports
- The live app can export derived gaze, pupil, and capture-record CSVs plus a JSON analysis report; its default workflow does not save raw eye video or persisted eye-crop images. source
- Gaze and events
- It computes gaze in degrees of visual angle and detects fixations, saccades, and microsaccades using I-VT, I-DT, and Engbert–Kliegl methods. source
- Inputs and outputs
- The CLI can analyze gaze CSV files and export JSON reports, and can record gaze and pupil CSV data from a camera. source
- Install options
- The README documents Python installation through uv, with capture, dashboard, figures, and web components available as optional extras. source
- Integrations
- The project lists OpenCV, MediaPipe, FastAPI, Uvicorn, Streamlit, Plotly, and Matplotlib as optional dependencies. source
- Intended audience
- The package metadata identifies science and research as its intended audience. source
- License
- The project is released under the MIT license. source
- License
- The software is released under the MIT License. source
- Live interfaces
- Optional interfaces include a local FastAPI and WebSocket HTML app and a Streamlit dashboard. source
- Pupil analysis
- Pupil tools include blink detection and interpolation, preprocessing, and eye-crop segmentation that reports pixel or pupil-to-iris-relative size. source
- Purpose
- iTrace is an open-source webcam eye-tracking analysis toolkit with a NumPy/SciPy analysis core. source
- Purpose
- iTrace is a webcam eye-movement analysis toolkit for gaze, saccades, microsaccades, and pupil measurements. source
- Security and privacy
- The project describes its default export as privacy-preserving and says live viewing may send a transient JPEG eye crop to the browser over a local WebSocket. source
- Use case
- The README presents it as a reference implementation of eye-movement estimation algorithms and advises comparison with a reference device or public dataset before reporting device-level accuracy. source
- Validation limit
- The project says it is algorithmically verified against synthetic ground truth but has not been validated against real eye-tracking data or a reference eye tracker. source
- Validation limitation
- The README says its synthetic and closed-loop checks are not real-device validation and that no bundled real-device validation is claimed. source
Checked against the maker's pages on 10 Oct 2026.