A Deep Learning Powered Eye Tracker

Pupil Labs pioneered deep learning powered mobile eye tracking. Currently the only company that offers a highly accurate and robust eye tracking system that works without any calibration or setup.

Research-Grade Eye Tracking Data

Neon has two eye cameras, one for each eye, running at 200Hz. The cameras are optimally located near the nose to ensure maximum visibility of the eyes across all subjects and minimizing occlusions. You will always have access to all the raw, intermediate, and final data - no special licenses required.

Video

Gaze

Fixations

IMU

Audio

Blinks

Data Streams

Video

Gaze

Fixations

IMU

Audio

Blinks

Coming Soon

Eye State

Full 3D pose of each eye in absolute units. Expected in Q3 2023 in Pupil Cloud.

Pupil Diameter

Available later in Q3 2023 in Pupil Cloud.

Neon Eye Tracking Module

Small package, big capabilities. High speed eye cameras, wide angle scene camera, stereo microphones, and IMU all encased in water resistant silicone.

Scene CameraEye CamerasData and Power InterfaceMicrophoneIMU

Eye Tracking for the Real World

Every Environment

Neon's gaze estimation pipeline is uniquely robust. It works with equal quality in any environment from complete darkness in a lab to a bright sunny day outside.

Sun & SurfNight NavigationScreen Time

Every Activity

Neon has been trained to handle slippage. Neon provides a stable eye tracking signal even during highly dynamic activities. Running, jumping and fast head movements - no problem.

Playing TennisRock ClimbingHiking

Every Subject

Neon works with every subject. NeonNet has been trained with a diverse range of eye colors, skin colors, face geometries, eye makeup, and contact lenses. We also offer a wide variety of frames for different fits and research needs - we’ve even got frames for kids!

Hardware Integration

The Neon module’s form factor makes it easy to integrate into all kinds of wearables and custom hardware. The interface geometry is open so that you can integrate Neon into your own prototypes. The required hardware interfaces are also open so you can read out the raw sensor data yourself.

Hardware Integration