Hi @user-185d95 👋 ! I had a look at your recording, and I believe that the issue may not stem from the fixation detector itself, but rather with the underlying pupil detection data.

HMD Eyes library in Pupil Capture includes an additional gaze pipeline that uses the pye3d model-based approach. This pipeline outputs gaze data translated into VR coordinates rather than the scene camera but relies on the 3D detector and proper pupil detection confidence.

In your recording, there are several moments where the 3D pupil detection either fails or jumps (see screenshots). These detection issues introduce noise and jitter into the gaze signal, which in turn impacts the quality of the derived fixations.

When using HMD Eyes (Core Add-ons for VR), the camera placement most of the times can't be optimized for data capture, so it's important to fine-tune the detection settings. Here are few things you can try:

• Ask the subject to roll their eyes in all directions and freeze the model once tracking looks stable, this should prevent some jumps.
• In the pupil detection settings (eye window), set the min/max pupil size to reflect the small pupils you're seeing, otherwise it would fail when pupil occupies less pixels than the minimum set-up.
• Switch the capture mode to ROI (Region of Interest), and define a small region where the pupil is expected.
• Adjust the pupil intensity parameters

These adjustments should reduce noise in the pupil signal, which will improve gaze accuracy during calibration—and ultimately result in more reliable fixations.

Even if you do enable it in the software, the VR-Addons do not have a scene camera. By enabling and selecting an eye camera you might end up with issues like that one.

So if I understand correctly, you're using a VR Add-on but prefer not to use Unity — is that right?

While this is possible, it is more involved and does come with far more complexity.

Calibration in Pupil Capture involves identifying calibration targets and correlating their positions with the pupil's position. Typically, this is done in the scene camera coordinates, which is why the standard Pupil Core setup requires a scene camera.

However, the VR Add-ons don’t include a scene camera. Instead, they rely on a special calibration routine that uses target positions VR coordinate space. Our Unity library handles the presentation of these stimuli and forwards their position to Unity for you.

If you want to skip Unity, you'll need to build a custom calibration method that replicates this process — essentially handling target presentation and coordinate mapping on your own.

and why it had these connecting reconnecting problems?

You are having this issue because you enabled "Scene/World camera" in the settings and selected one of the eye cameras. Now both the scene camera window and the eye camera are trying to access the streams at the same time. Simply disable the world camera.

Sure, you should be able to select Activate camera,, but if not, could you click on the Gear icon (General Settings) and "Restart with defaults"?

For pupil size alone, you do not need to calibrate.

Hi @user-c39646 , I'll respond point-by-point:

• When it "saves" (i.e., when the recording stops), it actually was doing everything from the Pupil Capture interface; not just "as if" it were doing everything from that interface. It is like a remote control.
• In programming, the clean up stage is usually where you close resources that are open and free up used memory. It is a recommended practice for leaving your computer in a well-maintained state, so that other programs are not disturbed.
• Yes, the script does work from another computer. You need to change the IP, as you say, and yes, they should be connected to the same local network.

I'd recommend reviewing this part of the Documentation for more clarification.