Is it possible yet to use to Neon glasses as if it was a Pupil Core headset with a MOS rather than with MacOS and Linux as there had been a mention a while back of developing this facility and our eyetracking analysis software is developed for MOS so it would save us the hassle and expense of requiring one device to capture the data and then connected to another device to analyse the data with our software?

We want to do eye tracking for patients with ALS, MS or similar diseases, in cheap or free. We want to map the gaze to the screen. Do you know an easier way than the following:

• get an IR-camera and a world camera.
• start Pupil capture and set it up.
• activate the surface plugin.
• put april tags on my screen.
• do calibration.
• then grab the surface data via UDP for further processing. Since the surface may not perfectly reflect the screen, more calibration is needed.

I'm asking because I was not able to resize the surface in Capture unfortunately, and I heard your software to Invisible works with less calibration steps.

Any help is appreciated.

Also, as granted by your documentation page https://docs.pupil-labs.com/core/, section Chat, I hereby say "hi".

Also any news on what conferences you will be this year, which are in germany?

Hi @user-84387e ! 👋 Great project!

That is indeed the process with Pupil Capture. However, can you provide more detail with respect to "not able to resize the surface in Capture"?

Since you mentioned Pupil Invisible, have you seen our latest eyetracker, Neon? It is also calibration-free and an improvement over Pupil Invisible in every way. If you want to map gaze to the screen in real-time with it, then AprilTags are still an option, and the setup for that is also simpler.

In my Logitech c615 it is uvc compatible but yet it is not giving me pupil capture in world feed

Hi @user-84387e , you are welcome.

If it's alright, I'll respond point-by-point:

• Surface adjustment: In Pupil Capture, after defining a Surface, there will then be a pink dot next to one of the corners that says edit surface, as shown in the attached image. If you click that pink dot, then you can drag the four transparent dots that correspond to the corners of the Surface.
• If this is not what you are referring to, then please let us know. Feel free to also share an example recording with [email removed] if that would be easier.
• Pupil Core & Neon: Understood! So, Pupil Core is not compatible with Neon software, but principles & techniques do carry over. To be clear, if you use Neon with Pupil Capture, then you will actually not use Neon's deep-learning powered, calibration-free pipeline. You would be using it as if it were a Pupil Core. Also, Neon support in Pupil Capture was experimental.
• Pupil Core based mouse cursor: This is possible also with Pupil Core. The source code in that Gaze Contingent guide is open-source and could be adapted. The key parts are having a real-time gaze signal and AprilTags to map gaze to a Surface.
• There are also members of the Pupil Core community who have made such tools <- that one also demonstrates self-made AprilTag clips, which your users might find helpful.
• Conferences: Thank you for the Frankfurt recommendation! I look forward to when we do get a chance to meet!

We are doing an experiment that require participant to stare at stimulus on monitor that is 300 cm away from them, is it possible to do calibration? If possible, should we still be concern about the accuracy of data collected? We are collecting gaze position and pupil diameter

In addition, I notice the gaze norm position is generated based on pupil norm position, is it possible to get mapped gaze position from the two pupil's norm position separately, without having to covert them to enforces monocular gaze estimation?

Thank you in advance!

yes, it will be nice to get gaze data sample from monocular base-data, thanks for recommending the plugin. Does the plugin affect the IPC backbone, essentially, after installment, whenver we subscribe to gaze. through ZMQ, it is going to return gaze data on monocular data of each side

hey, me and my partner are writing our bachelor thesis using the pupil Core, in our project our final goal is to be able to play a game like chess running in python only using eye-tracking. At the moment we are having trouble with the accuracy of the gaze, it is very jumpy and a bit laggy. we have setup at surface on a monitor inside the pupil capture software, then using a python script(better_eyetracking.py) using the API we get the surface coordinates and sends them over a socket to our simple test python script(simple_test.py), where we map the “norm_pos” from the surface to our python script by simply multiplying each coordinates by the height and the width of the test scripts window size.
We have looked a bit into smoothing the eye tracking using 1EUR filter. But we wanted to make sure that we are not missing a smarter way of mapping the surface coordinates to our test script(simple_test.py), which is running in pygame. My questions are therefore; is there a smarter way to map the coordinates from the surface? Is it a bottleneck when the better_eyetracking.py sends the gaze data over a socket to simple_test.py?

This is out first time working with eye-tracking, I hope the question makes sense. Thanks in advance!

Can Pupil Core handle multiple Cameras with the same name? If so, how? For us, it gets confused and always picks the first one.

Can you provide a guide while detection of pupils there are many lines light blue , blue, red near the pupils

1. How they help in pupil detection
2. How to tweak them to unlock maximum potential of pupil detection

thanks, nice to know that we are on the right track! but what do you mean by UDP?

Hello

I want to ask why I am having only one eye window open in Capture even though both are enabled?

Hi @user-d5bcb0 , would you be willing to share an example recording with [email removed] Then, we can provide direct feedback.

You can also send a screenshot of the Pupil Capture interface here, if you prefer. If AprilTag placement could be improved, for example, then the current setup could be resulting in inconsistent gaze mapping during Surface Tracking.

Otherwise, if the two scripts are running on the same computer and there is no computationally intensive routines that block the gaze sampling, then the data transmission should be consistent.

And, always great to see users helping users, @user-84387e 😊 Just to clarify and if it helps, while the Pupil Time Sync plugin uses UDP, the Network API uses ZMQ in TCP mode, where ZMQ is brokerless by design. While you can build a message broker system on top of ZMQ, the Network API uses the PUB-SUB pattern.

Yep, nice when users give false-ish information to others 😄

To the DIYers: What world camera would you suggest?

@user-84387e , do you mean they register with the same name to UVC libraries? That is, do you see them listed as separate devices when running this example from pyuvc?

You might just need to change this line in Pupil Capture for your purposes, and it might require an additional change to the downstream code that depends on devices_by_name.

Is there any way to set the size of the calibration window? When I use the screen marker mode I am able to disable full screen calibration, but then the window is very tiny. Is there a way to control the size of it?

Hi, I’m currently working on psychopy with the pupil core eye tracker and was wondering if any of you has been successful when trying to connect the 2 and gather data. I have followed the documentation provided, downloaded the necessary plugins and calibrated the eyetracker but I haven’t been able to successfully use the eye tracking during experiments.

Thanks, what do you mean by surface name ?

Currently I have psychopy_iohub_surface which is the default

Hi @user-d5bcb0 , thanks, we received it and I have taken a look. Although the calibration phase was not included in the recording, the gaze signal that you have looks qualitatively quite good. The 3D eye model has been fit well, the pupil is well centered, and pupil detection confidence is high throughout the entire recording. The Surface Tracking also looks to be working well.

So, to clarify, is this a recording that exhibits the "very jumpy" data that you mentioned in your original post?

Some other questions as a follow-up:

• May I ask why you do not use both eye cameras? You will generally get more accuracy that way and it seems your setup can accommodate that without issues.
• It is hard from the recording alone to see the lag that you referred to. Do you have some external reference measurement of the lag? For example, as done with these two scripts.

And, a tip:

• You can probably make the AprilTags smaller to save some screen real-estate.

Hi, we are using the LSL Lab Recorder to combine EEG, Psychtoolbox Markers and Pupil Core. In our pilot data we observed a few thinks I'd like to hear your opinion on: - In the 'time_stamps' column of the Lab Recorder output xdf of the pupil core stream are negative time stamps. We were wondering if that's because of the UDP package exchange, i.e. they are usually correct and we can just resort the time stamps and use them then. - Second, we notices that what the Lab Recorder saves as e.g. 'gaze_normal1_y' differs gravely from what is saved under that name in the gaze.csv. How can you explain that and is there a way that makes sure the data is equivalent? After all, synchronization is useless if the data ín the recorded via the LSL is not usable.

Thanks a lot for your support!

Hi @user-17e555 , may I first briefly ask why you want to change the calibration window size, in particular to make it smaller?

Hi @user-138bf5 , may I ask:

• How are you loading the XDF file?
• How have you temoprally aligned the LSL and native Pupil Capture data?

Hi, I have a MacBook and opening the Pupil Capture, the cameras do not shown any image and en error appears. I have tried then pupil core in an ubuntu and they work. How could I solve this issue with Mac? it it due to the USB?

Hello, does anyone know if there are software or plugins that could put April Tags on the PC screen on top of any other software? Thanks!

Probably any modern windowing/widget toolkit. I use Qt, but prefer coding in Python, so I use the official Python bindings for Qt called PySide.

Although the gaze-controlled-cursor-demo project is made to work with Neon, you can see an example of how to use PySide to overlay tags on top of other windows in the UI code

Hi @user-d5bcb0 , it would actually be helpful to see one of the worse recordings, in particular one that includes a recording of the 3D eye model fitting procedure, as well as the calibration routine, if you have that.

With respect to the lag issue, multiplying norm_pos_x/y by the screen resolution is a fast operation, especially on modern systems. If you are regularly seeing lag, then the issue is probably somewhere else.

And, you are welcome!

Hi, I am currently working on an academic publication in which I use Pupil Core eye-tracking glasses. The reviewer is asking for an image of the glasses to be added as a Figure. Do you have such an image and can I get permission to use it for publication? I will include the source and include appropriate credit of course. Please let me know how to best proceed.

Hello, I’ve always been a satisfied user of your Pupillabs Core product. However, I have a few questions.

It is stated that the 3D gaze mapping in the Core product exhibits an accuracy of 1.5–2.5 degrees. I’m wondering how this accuracy (or error) was calculated. I’d also like to measure how the accuracy changes as the viewing distance gradually increases.

Is the claimed 1.5–2.5 degrees of accuracy derived by taking into account the sensor size, resolution, focal length, and field of view (FoV) for both the eye camera and the world camera? If so, could you please share the formula or any related academic references/papers?

I look forward to your kind response. Thank you.

Hi! I have a problem with opening the pupil capture and player on my ubuntu. I was able to use it some months ago but now the programs do not open and if IU try to open them using the terminal, I get the following error: Unhandled SIGSEGV: A segmentation fault occurred. This probably ocurred because a complied module has a bug in it and is not properly wrapped with sig_on(), sig_off(). Please find attached the whole error. Any idea on how to solve it?

That option does not appear..... @user-f43a29 ; it used to appear some months ago but it doesn't appear anymore.

Hi Pupil Labs,

Please some world IDs are missing in the gaze_position_on_surface. How can I get all that in the export

Look at the jump from 15898 to I6035 and then ,to ,17242

Nope. Here is what I did; I keep exporting the same data over and over again and those missing samples came up. The participant looked at the surface and never looked aways

Hi, I currently own, in my lab, the old version of Pupil Core which only has one eye camera. I am wondering what would be the advantages of getting the new version of Pupil Core ? Is the data quality better ? Thanks in advance for your advice !

@user-f43a29 @user-d407c1 , one eye camera of Pupil Core is switching on and off ... what could be the reason ?

Hi all, my lab is using the Pupil Labs Core and we were looking at the gaze positioning file exported from our recordings. We were hoping to understand how the gaze normal values are calibrated with the device? Does it utilize and eye model using the visual axis or the optical axis? Likewise is it different for both the gaze point and gaze normal data? Thanks!

Hi, I am using pupil core on two different windows device, I notice on one device gaze data receive from network API has a b’ prefix (which I think mean byte string), the other one don’t, is there a setting I can toggle or flag I can set to make them align? Thanks

In addition, I received halve the sampling amount within the same amount of time when attempt to read using networking API

I know that Pupil Mobile has been deprecated for the Core system, but I wonder what people are doing if they need that capability for this system? I have a couple of Cores that I'd like to use in a student lab for mobile eyetracking, but I'm not sure that capability still exists. Any thoughts and pointers welcome, although I don't have much in the way of funding for this project.

Hi all, I'm trying to calculate the distance between the calculated gaze positions and my reference targets, but I'm noticing that the normalized gaze position is on a [0,1] coordinate plane whereas the reference targets seem to follow a different coordinate plane (seen in reference_locations.msgpack). Anyone know how to reconcile this difference, or if there's a better way to calculate this distance?

Hi, I'm using the Pupil Core in my project to gather users' gaze positions and fixation data while navigating a Unity virtual gallery. The data collection is based on a monitor rather than a head-mounted XR device. Currently, I can export the data from Pupil Player, but I still can't correlate the Pupil data with the Unity World Space coordinate system to find the exact object the user looked at though I utilized Surface Tracking by adding markers to the corners of my monitor. How can I resolve this? BTW, how should I start if I want to enable real-time Pupil data transmission using the Network API? Thank you

hello, is anyone run pupil core in aarch64 devices? I clone pupil repo in my arrch64 device, jetson nano or raspberry pi , it's so difficult to build wheel to install

Hi @user-da1455 ! We do not have native support for ARM processors, you could try to run it from source and compile wheels, but to my knowledge, there are some additional dependencies which do not support ARM processors either, so, you would need to compile them too. Please have a look at https://discord.com/channels/285728493612957698/1039477832440631366

May I ask why would you like to run it on an ARM processor? If all of this sounds to complicate and you simply want some single board computer (SBC) to run it, some users have reported that the Lattepanda 3 delta works great, and without headaches as it's architecture is x86. https://discord.com/channels/285728493612957698/285728493612957698/1123484310649966602

Thanks for you reply. About you question, all I can tell you is that we have a project that requires a board with the aarch64 architecture, and there is no way to replace it at this time, so we need and try to do our effort to run the pupil core. We attempt to run from source . But meet to much problems, well, we will try again

okay, thank you for the quick answer, I

Forgive me to ask one stupid question: Is that pupil service/capture software auto publish camera video stream via http network that can be subscribe by other program such opencv or website?

I use the python realtime api but cannot find any devices>

https://github.com/pupil-labs/realtime-python-api I use this repo

Hi @user-da1455 ! The API that you link is for 👓 neon / 🕶 invisible , for Pupil Core please refer to https://docs.pupil-labs.com/core/developer/network-api/ which uses ZMQ and msgpack.

Hi everyone,

I’m using a Pupil Core device on a Mac. The first time I connected the device, everything worked fine – the software detected the cameras and I could record video normally.

However, when pupil capture software froze, I had to force shut down and restart my computer. After that, the software still opens, but it no longer shows any video feed from the cameras.

I’ve tried reconnecting the device and restarting the software, but it doesn’t help. It cannot connet the device.

Interestingly, this also happened once before, and after I left the device unplugged for about two weeks and tried again, it suddenly worked again.

Has anyone experienced something similar?

Thanks in advance!

[email removed] thank you for your support! Now

Hi，How to set Pupil Core Scene Camera to 120Hz mode? Can it actually achieve 120Hz?Thank you

Hi everyone, I’ m using a Pupil Core device on windows and I have some issue with the calibration. Could you please help me?

Hi Rob !I’m trying to calibrate the Pupil Labs eye-tracking glasses, but I can’t properly calibrate the surface. As a result, I can’t extract accurate data or generate maps

Hi! The core has been great so far, but I've run into an issue with the eye cameras. The right eye stopped connecting, so I tried to plug it on the left eye's cable to see if it was a camera or cable issue, now neither works! No wires pulled out of the connector or anything I can see. In device manager (on Windows) I see 3 "Pupil Cam" devices under "libusbK USB Devices". If I disconnect one of the eye cameras only two devices show up there. So it seems the connection is ok, but I'm not getting any eye cameras in Pupil Capture. Any ideas?

Hi there,

I’m working with Pupil Core eye trackers alongside a RealSense camera, but we’re experiencing a significant drop in the RealSense frame rate after a few seconds. Adjusting the resolution or frame rate of the world camera in Pupil Capture temporarily restores the RealSense frame rate (regardless of the setting), but it quickly plummets again.

The system runs on an Intel 11th-gen i7-11800H (2.3 GHz, 16 cores). Pupil Capture reports a CPU usage of 200, yet system monitoring shows all cores at only 20%. We're using Pupil Capture version 1.21.0 due to compatibility requirements with legacy code, so upgrading isn't an option.

Do you have any suggestions for resolving this issue without upgrading Pupil Capture? I realize this is a tricky constraint, but any insights would be greatly appreciated!