Hello, I had a question about the storage. In my cloud data it says I have a full database of 2 hours, but when I go to delete videos it still continues to tell me that I have full storage and that I need to manage it. I am not sure what is going on.

Hi @user-7e49f7 , have you also deleted these recordings from the Trash? See attached image.

Are questions about neon player okay here?

So I was downloading the video from neon player but I can see that my gaze data is completely empty and I was wondering what could be causing this?

I click to the left side of the neon player with the download symbol.

Yes, I am.

Did you need the video after I downloaded it from neon player?

I mean the original recording that you load into Neon Player, not the exports.

What do I need to do when my reference image gives an error. I tried other images of the environment but none of them is working. Can anyone help me

Just shared the video please let me know if there are any issues.

Hey @user-7e49f7 , thanks. We would need the full recording folder though to investigate, not just the scene camera video.

However, glad to hear that you resolved the issue!

Hello, we have an offline Windows 10 machine where we installed PsychoPy 2024, but we were unable to install (lots of network errors) the Pupil package using PyPI (https://pypi.org/project/psychopy-eyetracker-pupil-labs/#history). Is there a way to install the pupil lab package without internet access? Also, is there a list of all the necessary dependencies?

Hi everyone! I’m working on synchronizing multiple data streams for a recording session, and among those streams I have three Pupil Neon devices. I’m currently using LabRecorder (LSL app) and successfully recording synchronized data from all three Neons into a single XDF file. However, I’d also like to synchronize the data stored on Pupil Cloud for the three devices. I’ve read about lsl-relay, but I’m wondering whether it’s effective when working with three Neon devices simultaneously. Is there a recommended way to handle this kind of multi-device synchronization between LSL recordings and Pupil Cloud data? Thanks a lot for any advice!

Hi @user-b407ae , the lsl-relay is deprecated in favor of the integarted LSL functionality in the Neon Companion app.

If you want to synchronize the Timeseries CSV data from Pupil Cloud, then you only need to send some Events to Neon during the recording, perhaps one every few seconds. These Events will be saved in Neon's recording timeline and LSL's. Then, after you have downloaded the Timeseries CSV data, you simply align the Events with their corresponding entries in the LSL stream and you have synchronized.

Hello! I want to export the gaze data (fixations, saccades, blinks, etc.). I have two methods, either using pl-rec-export or from Pupil Cloud. But since Pupil Cloud creates additional grey frames at the beginning and end of the main data, and I don't prefer using it, because later on, I need to extract frames from the original raw video exported from the phone (which does not contain grey frames). When reading the caveats of the pl rec export function, they said: "Fixations and saccades detected locally might vary from those exported from Pupil Cloud. This is due to differences in real-time gaze inference when compared to the full 200 Hz gaze data available in the Cloud." After comparing, I found that the number of fixations and saccades slightly differ, along with their characteristics. My question is: Despite this difference in detection, can I use only data from pl rec export and consider it as reliable 100%? Because my main interest is to keep using the raw video file exported from the phone, not the one with grey frames from Pupil Cloud.

Yes. That's correct. I used this code.

It worked out - I meant to say that the enrichment can detect the tags for some frames, but not for others. So I thought the implementation of the tags should be correct.

Thx a lot. I have invited this account to the workspace! My account is [email removed]

For 3Match and newObject projects, I encountered the issue with the detection of tags for some frames. The program was built in jsPsych with AprilTags.

Hi @user-5be4bb , the pl-neon-recording library is now preferred for working with the raw data at the command-line or programmatically, rather than pl-rec-export. You can also use Neon Player to export the raw data to CSV.

The pl-rec-export library was reliable, but on the latest version of the Neon Companion app, fixations and blinks are computed in real-time and saved on the phone. Saccades are taken as the inter-fixation intervals. Aside from that, the pl-neon-recording library exposes a more ergonomic programming interface.

Please note though that gray frames do not interfere with analysis. You can simply skip over them. Otherwise, the Timeseries CSV data is in principle the same data as saved in the Native Recording Data, but in a more user-friendly format, and re-processed at 200Hz.

Hello, I performed an acquisition with the Pupil Labs Neon. The recording on the cloud is fine, and I downloaded the “Native data recording,” but when I drag the folder into the Neon Player it opens briefly and then the program closes. I tried downloading the folder again, but the same problem persists. Could you tell me why this is happening? Thank you very much!

Hi @user-a4aa71 , could you provide the logs in the neon_player_settings folder? This is found in the user directory of your computer. You can send it via DM, if you prefer.

https://discord.com/channels/285728493612957698/733230031228370956/1385099425680064562

Hi Pupil Labs, We're trying out the new GPS app but when we click the white button to start recording, it turns to gray and displays the message "Initializing sensor ..." The message doesn't go away and if we click the recording button again, it displays the filepath where gps csv is stored. All the stored csvs are however blank. We've allowed location and other settings and the phone is connected to internet. What could be going wrong here?

I used the library from here to create visual tag patterns that look like AprilTags from this family (AprilTagFamily(tagConfig36h11)). I used the <script [email removed]

For the event.csv, is there a way that I could define each trial with a unique name to send to the eye tracker? Currently, it does not indicate the unique trial name. I am using the plEvent component of the pupillab's plugin.

Hi @user-9a1aed , similar to how my colleague @user-d407c1 explained here (https://discord.com/channels/285728493612957698/1047111711230009405/1385566090607984660), this is possible in PsychoPy. It requires the use of their variables and conditions files. Since this is more about creating a custom string in the context of PsychoPy's system and not specific to the Neon PsychoPy plugin, their Support team might be better positioned to help you with this request.

Hi, I'm currently trying to enrich my recordings on the Pupil Cloud using a reference image of the environment, but I keep encountering an error that prevents me from proceeding. The error message I receive is: "The enrichment could not be computed based on the selected scanning video and reference image. Please refer to the setup instructions and try again with a different video or image."

All my recordings are successfully uploaded and visible on the Pupil Cloud. I've tried using multiple reference images — including screenshots taken directly from the video and separate photos of the environment — but none of them work. I also attempted the enrichment with different scanning videos and different timestamps within those videos, but the same error keeps appearing. Could you help me?

For the neon eye-tracker, when the white LED is on, what does it indicate?

I am using the eye-tracker today, and the LED was on. After the recording session, I looked at the recording, and there was no fixation information (red circle) on the video. Does this mean the eye-tracking data is lost?

My tutor Marieke Otterdijk has invited you.

Have you received another invite?

I’ve waited for upto 2 minutes but the initializing sensor sign doesn’t change

Hello! I was hoping to get some advice on how to best use events. Each of my recordings contain 16 events with unique labels to preserve their order over the course of the task (e.g., Event 1, Event 2...). I then run the Reference Image Mapper on the full recording and draw AOIs. My goal is an output file with the fixations to each AOI during each event. However, I noticed the event labels are not included when I download the Reference Image Mapper report. I understand one solution would be to run a separate enrichment for each event, however this would be quite time intensive given the large number of events and recordings. What is the best way to obtain an output file with fixations to the AOIs in each event?

Hi @user-13d297 , sure! All Enriched data is on the same time axis as the rest of Neon's standard data. So, you can still run the Reference Image Mapper on the full recording, if you prefer that approach. Then, use the timestamps in the events.csv file in the Timeseries Data download to filter and choose subsections of the AOI fixation data to analyze or export to a separate file.

Hey you guys! Our lab just got our pair of Neon glasses and we’re testing them out today. We’re super excited about them! Just a quick question: do other people notice that they are a bit hot to the touch after using them for prolonged periods of time? After 20min or so they’re getting pretty warm. We plan to run experiments that are an hour or more. We’re curious about others’ experiences and how they might address this in their labs. Thanks so much!

Hi @user-ffef53 , are you referring to the Neon module or the phone? The module is expected to warm up during normal use and the phone can also warm up. Please see this message for more details, also if you'd like foam pads for your frame: https://discord.com/channels/285728493612957698/1047111711230009405/1383005671611433083

Hi Rob, thx a lot for your reply. The video captures a PsychoPy program built with the PupilLab's plugin functions. I do not think it has anything to do with the Python-C++ package.

The recordings on Pupil Cloud, yes, the Tags changed very often since participants responded quickly to the visual stimuli. But this program is built with jsPsych, is there a way I can send Events to mark the start of the trials to the Neon eye tracker? If there is, is it possible that you guys have a demo jsPsych program? Thx

Nope. It is a standalone jsPsych program. I added the Tags as images into the program around the four corners.

The recordings on Pupil Cloud showing that some of the tags are detected while others are not --> is built using jsPsych.

We do not have demo jsPsych programs, but the principles of Events remain the same. You send them as in an HTTP Post request to the /api/event endpoint of Neon's Real-time API.

Since jsPsych is a Javascript framework, it is quite easy to send HTTP Post requests from there and standard resources can provide the steps.

Ok, thanks for clarifying. Then, would you be willing to share the psyexp file for that recording with us at [email removed] Because the discrepancy in your video between the gaze circle in the app and the real-time gaze cursor in the PsychoPy program is not expected, unless there was lag or transmission delay.

I am connecting the program and the Neon eye tracker using Wifi. Not a hub. Will it cause the delay?

Thank You @user-cdcab0 , I have few more questions about the offset.

I am running studies where participants do not have to interact through gaze. So, I am planning to just adjust the offset manually after the recording using the Gaze Offset from Neon Player.

Two questions: - Am I correct that there is no problem in NOT applying the offset from the NeonCompanion app before the recording if the participant does not have to interact through gaze? - Would it be possible to apply multiple offset within the same recording? At the moment NeonPlayer allow to set the offset only once. However, in case of long recordings, say 30minutes, it would be great to have the possibility to change the offset if needed. For example, correct X 0.01 Y 0.01 from time 0 till 15min30sec and from 15min31sec correct X 0.00 Y 0.02. This would allow to have a single long recording and, by inserting regular checks within the 30 minutes recording, we can check the accuracy/reliability throughout the recording.

Thanks

Hello @user-f43a29 , Yes, Neon is calibration free. however, for both on screen and real life studies, it would be nice to do the following: - before starting the study ask the participant (ppt) to look at a spot placed at a distance of interest for the study - run the study - at the end, ask again to look at the spot as before. This can ensure any reviewer (and experimenter too) that data from start to end were reliable (e.g., did not drifted somewhere for any reason)

While I get the use of the wearer option, I can assure you that spending some time checking each recording is well-spent time and a necessary step to check data quality.

Yes, gaze offset is not a calibration, my bad, but as said above, being able to check say, every 10 minutes (e.g., during a break) that the same offset set at time 0 is still valid would be agreat add ofr the NeonPlayer (like you can add multiple surfaces, you may add multiple offset at different timepoints).

Hi @user-b407ae , you are welcome!

If the goal is simply to align the Neon and LSL streams for sync purposes, then you can use the basic Timestamp on arrival method.

And the second link directs to the Event Data Outlet section of our LSL Integration Documentation. That section is brief and shows at the top of the browser, just above Connection problems? as in the attached image.

Thank you this is very helpful. For the time sync solution, would I have to sent events through Neon? I currently have events/marks sent through LSL from Unity. Would I need to use Neon events to align with the other data (IMU, etc.) though? Also, is there more detailed information about time delays due to software/hardware implementation of LSL? I'm trying to figure out where inaccuracies could occur to make sure I account for them since I need millisecond precision. I'm using LabRecorder to record the data streams.

Perfect, thank you! Yeah, I have an LSL outlet for Unity events that I also record in LabRecorder so I can align that with the Neon LSL stream. I mainly wanted to check where time delays may arise with Neon and LSL; it sounds like everything would be handled with LSL though. Would there be time delays with the Neon Events before they are sent to LSL?

Hi, I am using QR markers for years, but I was wondering whether it is possible to put the QR markers in the video in pupil cloud after the gaze measurements.

Hi @user-3c26e4 , do you mean edit the videos digitally to post-hoc insert "virtual" QR codes? If so, then this is not possible on Pupil Cloud. The QR codes must be physically present in the scene, at the time of recording.

If you are looking for a markerless approach, then we would recommend the Reference Image Mapper.

In one of our experiments using the pupillabs neon device, when checking the data collected in pupilcloud we observed that sometimes blinks where not detected for some participants while for others there was an ! error and no pupil diameter was calculated...Any ideas of what happened?

Hi, at the moment I'm looking to only post-hoc sync the Neon and LSL timelines. Would other use cases be to mark important moments through Neon Events? I'm using Unity to send signals to all connected devices (Raspberry Pi which is connected to an LED for the glasses--but trying to replace this with LSL, motion tracking, FNIRS, etc) in order to be able to post-hoc sync everything. I have taken a look through the tutorials and made the Unity-LSL connection. That's good to know about the delays; I'm still in the process of verifying the integration/precision.

Hi, We're conducting a study using the Neon eye tracking glasses and are collecting an hours worth of data per participant. However, we've reached the 2-hour Pupil Cloud storage cap and are looking for alternative ways to manage and store our recordings without losing the data we want to analyse. Any suggestions on the best way to mange this issue would be great.

Hello! Just received the neon glasses (very cool), have done a few recordings and it seems well. Unsure if I'm missing something so thought I'd come onto this and asked, but once exported the videos, the pupil-tracker (the lil circle on the screen itself), is no longer there (still on the raw recording on the neon app). Only have played around with the pair for 30mins so still quite new to it!

Hi @user-03b75d , great to hear you are enjoying it!

Your device is working fine and what you report is expected. To get a video with the gaze circle, try the Video Renderer Enrichment of Pupil Cloud or the Video Exporter plugin of Neon Player.

If you have not yet had it, I recommend signing up for your free 30-minute Onboarding call. Just send an email to info@pupil-labs.com with the original Order ID.

Hi! I'm trying to download data from the cloud, but I'm experiencing unstable download speeds — they fluctuate a lot (from 0 up to around 65), even though my internet connection is stable. Because of this, the download fails. Could you please help me troubleshoot this or suggest what I can do to fix it?

Thanks in advance!

Hi! I'm trying to download data from the

Hi @user-f43a29 , thanks again for your helpful input. Do you happen to know how many one-hour recordings the Companion app can store at once? For our workflow, we're planning to hold onto recordings locally and upload them one or two at a time to Pupil Cloud for analysis. Once we've extracted the necessary data, we’d delete each recording to free up space. Any insights on storage limits or best practices would be much appreciated.

Hi Rob, thanks again. Just to clarify—I was referring to storing the recordings on the companion app on the phone, and then uploading them manually to the cloud, one or two at a time, directly from the phone. Essentially, we would turn off the auto upload feature and upload each recording to the cloud manually, whilst using the phone for storage.

Hi everyone,

I'm working on an exploratory study focusing on shopper experience in a supermarket using Pupil neon eye tracking glasses. The goal is to understand attention and engagement patterns (e.g., what products shoppers look at, dwell times, etc.).

However, I don’t have reference images or a static scene to map gaze data onto. The supermarket is a dynamic, complex environment, so traditional surface detection and reference image-based AOIs aren't feasible.

I'd really appreciate help with the following:

How can I analyze gaze data without reference images? Are there best practices for working with raw gaze or world camera video in dynamic environments? What metrics can I extract in this context? For example: heatmaps Fixation durations Dwell time in broad regions (e.g., shelf vs. signage) Gaze distribution over time Any recommended tools or workflows for post-processing or annotating gaze data manually or semi-automatically? If anyone has worked on similar mobile eye-tracking studies in naturalistic settings, I’d love to hear how you approached the analysis.

Thanks in advance!

Hi @user-67b98a! Thanks for your question. Could you elaborate more on which areas of the shopping environment you're investigating? You do mention shelf vs signage. For these, our reference image mapper enrichment will often work well. E.g. check out this supermarket shelving example: https://docs.pupil-labs.com/neon/pupil-cloud/enrichments/reference-image-mapper/#_3-supermarket-shelf

Hey @user-67b98a, I'm a fellow researcher. If you haven't come across them yet, I highly recommend the Fundamentals of Eye Tracking series of papers, written by experts especially for researchers who are just starting out with eye tracking. They write about choosing tools, for example, in this paper: https://link.springer.com/article/10.3758/s13428-024-02529-7 As a relative beginner with eye tracking, I have personally found this series extremely accessible and helpful, so perhaps you will too.

Hello Neon Team. Just to now if someone is having problems when dowloading timeseries CSV from neon cloud. I am experimenting really slow donwloads that don´´t come to an end

Hi Team, I received this error when I was using the April tag frame. From the demo program, the experiment is using Height. But for my other program, I use pixels. So I converted the size [w,h] of the tag frame from [2,2] to the window size in pixels [1360,768]. May know if I did the conversion incorrectly to cause this error? thx!

Hello, I am master student at Ulm university. We are conducting a research. We have one problem. When we are trying to do a enrichment and after when we try to donwload this enriched video. All of the other videos datas are also coming within specific videos excel. We tried to change wearer but didnt work. You can see the screenshot. I would be very happy if you could help us download the data of just the video we selected.

yes @user-f43a29

@user-f43a29 it worked thank you so much

Hi @user-67b98a. Thanks for providing more context about your study. I understand that the dynamic nature of navigating a supermarket presents an interesting challenge for analysis.

For future studies, I want to clarify that the Reference Image Mapper in Pupil Cloud is well-suited for such environments. The process involves taking static images + scanning recordings of the areas of interest, for example, specific shelves or even entire aisles. You would then upload these. Pupil Cloud will automatically map gaze data from all participant recordings onto these images whenever a participant views those specific areas, regardless of the route they take through the store. Here's an example of mapping gaze throughout an entire room. The same principles apply, but scaled up to the supermarket. This allows for powerful aggregated analysis across many participants.

Since you have already completed your data collection, the best approach would be to use the Manual Mapper enrichment in Pupil Cloud. This tool allows you to go through your existing recordings and manually define fixations on AOIs. While this is a more hands-on process, it will enable you to perform the analysis on your current dataset.

Hello, the following questions appeared during my recording, I would like to ask if my recording record can be found?

Hi there! Our lab just got the Neon glasses and we've been playing around with some mini experiments. I had a few questions for the community:

1. Since we do not have Pupil Cloud, we’re writing scripts to convert all the raw files into CSVs on our own. Does anyone have experience with this? I tried looking for examples, but I don’t see much in the discord about this (outside of using the cloud)
2. Related to this, in the Neon Player you can convert some files into CSV format, but I only see that it’s possible for raw gaze files. Can someone explain this further?
3. I’m seeing some interesting differences between the folders when I export from the companion device. I see 2 folders, Neon and Neon Export, that contain basically the same info, but Neon Export has 3 more files (gaze_right ps1.raw and .time and info.json). Additionally, when I upload the folders to the Neon Player, I generate a new folder within the original raw data folder which contains even more files including world timestamps and a gaze.csv that would be super useful to have from the start. Does anyone have any best practices for organizing all this data? I think the workflow is a bit unclear. Its ok for now when we’re just running little tests, but in the future I’d like to know the best way of managing all of this raw data and converting it to CSVs without losing or misaligning anything.

Thanks in advance for your help!

Hi @user-ffef53 👋 ! Welcome! All of that is already possible, let me clarify some aspects and walk you through the workflows:

• Transferring Recordings: The general workflow for exporting recordings is described here. Both the raw and the export folders would contain all the relevant information. Generally, we recommend exporting it and using the Neon Export folder to prevent affecting the normal app's operation, but if you are going to delete the data afterwards is also fine.

With the data already on your computer, you have different options to export the relevant CSV files.

1. Using Neon Player:

You can load your recordings into Neon Player, then, go to the Plugins menu, enable the streams/plugins you want to export (e.g. gaze, pupil size, eyeball position) and click Export to generate the corresponding CSV files.

The general workflow is described here.

Just keep in mind: Only the enabled plugins will be included in the export.

2: Programmatic CSV export

If you're working in Python, and want to programmatically export the data to CSV files, you can leverage our pl-neon-recording library.
Here you can find a full example on how to export the data onto CSV files.

Let us know if you have any questions or get stuck somewhere.

Hello. I am a PhD student collecting data from the Neon glasses in children for a continuous recording of ~20 min. I have plotted the pupil size (collected from the 3d_eye_states.csv file) but am struggling to interpret where the blinks are - I am wondering if you have any advice? (I can send picture of the plot/data if that is helpful - there are many downward spikes, but they vary in their depth and occur at very irregular time periods - so am not sure whether they are blinks or not)

Hi @user-f389a1 👋 ! You can find the blink data in the blinks.csv file.

By default, NeonNet would try to infer pupil size at all times, even during blinks. You'll typically see a rapid drop in pupil size, though it may not reach zero. Gaze data is also still reported during blinks, you can see the signal turn noisy, but we don’t automatically remove it.

This is intentional, so that you have full control over how you define a blink. For example, you might prefer to base it on a different eyelid aperture rather than rely on a hard cutoff. Lemme know if you need help filtering them out or plotting them!

May I ask how you're analyzing the data? Any programming language you are familiar with?

Conceptually, if you’re working with the timeseries format, you’ll want to load both the 3d_eye_states.csv and blinks.csv files.

• Use the timestamp [ns] column from 3d_eye_states.csv
• And the start timestamp [ns] / end timestamp [ns] columns from blinks.csv

To remove blink periods from your analysis, you'd filter out all rows in 3d_eye_states.csv where the timestamp falls between the start and end times of any blink.

Here you have a snippet that exemplifies that, plotting blinks over the left pupil size timeseries and filtering out sections of blinks from a folder containing those CSV files.

Whether to interpolate or not, is a decision that you would need to make based on your experiment needs and requirements.

Hello! I am a Phd student and our lab currently owns a pupil core and currently looking into upgrading to a Pupil Neon. My current concerns are as follows: 1. Does Neon require eye camera adjustments like the core? 2. Can we save videos without the Pupil Cloud for Neon?

Hi, I was able to start playing around with the CSVs I exported and just noticed that my fixations csv is empty! Its strange because I have a saccade csv. Both are supposed to be generated together from the plugins menu in the Neon Player. The video is about an hour so I'm certain I have fixation events in the recording, but I think something is happening upon export that's stopping the fixations from being loaded into a csv separately. Please advise. Thanks! (Edit: to add, even with new recordings when I try to export fixations it doesn't work. In the player it says there are no fixations available when I try to click through them)

I should have the most up to date of both because we just received the device a few weeks ago. I just tried again on a new, much shorter recording and it worked, but for the longer ones its not working so maybe its a bug?

I think the processing just takes much longer, but maybe there's a way to optimize it that im missing?

Thanks for your quick reply! Fixations in the latest version of the app are computed directly on the device, as long as you have “Compute fixations” enabled in the Companion App settings.

That means Neon Player shouldn't take long to load them, since it doesn’t need to process them post-hoc.

Could you double-check that you’re using Neon Player 5.x?

Hi PupilCloud Team, could you let me know the dimensions of the red circle indicating gaze in the cloud - either in pixel or field of view (degrees)? Thanks!

Hello! I work for a market research company, and in a project with Neon glasses we are having problems with two videos loading in the cloud, it says "processing", but they don't load, these videos can be played on the mobile. Do you know what I can do?

Hi, I must answer some questions to the "Datenschutzbeauftragte" (information security officer?) concerning the videos of the Iris/pupil. could they be identified to a special person? is the video "safe" - in aspects of the data-security and re-identification of a person? Is there any information about this issue? Maybe I could link to a document? Is it possible to remove the eyes-video without loosing information? Thanks in advance!

ello everyone,

I recently used the Neon to record multiple videos. When I checked the cloud later, I noticed that six of the videos were not uploaded. I tried to upload them manually, but after a few seconds, the upload fails each time.

Is there any way to successfully upload these videos to the cloud?

Hi Dom, I have some questions for Neon: (1) is this eye tracker able to totally eliminate the impacts of long eyelashes? (2) is the angle of the world camera adjustable?

I am using Psychopy 2025.1.1 for an eye tracking experiment with videos. I have downloaded the Pupil Labs Plugin. In settings, I select Pupil Labs (Neon) but it does not allow me to Modify the IP address and port. I don't see those options in the dropdown.

same problem with that version, so I installed a newer one.

Hi, has anyone been able to synch fNIRS aurora lsl triggers and the neon recording? My triggers arent showing up for the neon recording

And also, does PupilLab have any devices that can output the gaze data in real time? So that we can combine the gaze data with other real time processing code?

Hello. I am trying to connect monitor my data collection in real-time with the neon monitor app, but the link, http://neon.local:8080/, doesn't seem to be working. Both my companion device and my computer are connected to the same network. Could I get help troubleshooting?

Hello, I’m currently trying out the Pupil Labs Neon our research unit recently bought. I appear to have some problems with local streaming, that is, syncing my computer with the Neon.

I am currently on PsychoPy trying out your Builder demo, gaze_contingent_demo.psyexp. While I can get it to run when using MouseGaze as the eyetracker device, I am unable to do so using the Pupil Labs (Neon) plugin. I’ve done the following troubleshooting steps:

1. Updated all packages included in the Pupil Labs plugin
2. Checked companion address and port to ensure it is neon.local and 8080 respectively
3. Also tried the Coder .py file version, which didn’t work
4. Pinged the eyetracker via ping neon.local which did show a connection

Tldr; I would just like to try out the demo but unfortunately, it doesn’t work with my eyetracker. Is there any additional steps that I’m supposed to take?

Hi @user-f43a29 I am at home, using a home wifi. I believe it is a wifi 6

Hi @user-f43a29 , thanks for the quick reply. So my mobile hotspot did nothing to change it. I do realise that trying to access Neon Monitor via the URL, neon.local:8080 or the IP address provided on the Companion, results in some extremely slow lag, if not just screen freezing.

I’ve tried running dns-sd -B _services._dns-sd._udp via MacBook Terminal, and I couldn’t find anything related to the eyetracker. I do agree with you that it may be a router issue but my wifi and hotspot does allow for mDNS and UDP traffic. Any idea what I can do next?

there is a problem with the installation of the plugin. In the settings list only Pupil Labs appears but not Pupil Labs (Neon).

yes I did it... and I tried in PIP console of Psychopy 2024.2.4 pip install --force-reinstall --no-cache-dir psychopy-eyetracker-pupil-labs

Yes I can see it in preferences - startup plugins but it is not appearing in the list of downloaded packages. Pupil Labs plugin it's visibly installed in Plugins list...

Hi @user-4c21e5! I did try the troubleshooting steps. Both devices are connected on my university's wifi. Is that what you were asking?

Hello everyone, I’m having trouble uploading 7 videos to Pupil Cloud from the phone, as the uploads keep failing. I can extract the videos from the phone, so I was wondering if it’s possible to upload them to Pupil Cloud directly from a computer. This seems like my best option, since I’ve already contacted Pupil Labs support and fixing the issue on the phone appears to be difficult.

Hi, yes. It is working when we run other experiment (without neon) with other applications like psychopy but if we send triggers directly from aurora to the neon glasses it doesnt work. Thats why i am wondering about connection with the aurora app specifically if that makes sense

Thank you Rob, the deadline for my master thesis is in 30 days, so you can understand the pressure I am in right now.

@user-f43a29 Yes. We have now managed to make it work for one pair of neon glasses, but the experiment involves 2 pair of glasses recording at the same time. The api only seems to recognize one pair at a time

Hi @user-21cddf , can you describe your setup in more detail? It sounds like you rather receive Aurora‘s triggers programmatically and then convert those to Neon‘s Events in your own Python code?

Hi everyone, I have a very simple Python script that remotely controls the start and stop of recordings. The code uses these two functions: recording_id = device.recording_start() and device.recording_stop_and_save(). I connect my PC and the smartphone connected to the NEON to my phone's hotspot (the code searches the device by IP). The problem is that sometimes (very often actually) I get this error --> Error: Python Error: DeviceError: (400, 'Cannot stop recording, not recording') etc ... I noticed that if I keep sending start and stop commands, the recordings either don't get saved or are corrupted (they contain no video of the scene), even though no error appears in the Python console. Moreover, after this error occurs, it's impossible to get valid recordings anymore—I had to reboot the device. How can I solve this issue? Next time I run experiments, I will make sure to use a router, but is this a connection problem or is it caused by something else? I have already used my phone's hotspot, and with the Pupil Labs Core I have never had this problem. Thank you very much.

Hi, does anyone know of any publications or code bases that focus on gaze contingent research with multiple neon devices?

Hi @user-a4aa71 , this is not a connection issue. May I ask if your code also uses device.close() at the end of each session, and whenever it is quit, including when it is quit by Ctrl+C?

my code is this one, very simple... (I'm

Hey guys. I may be doing things a bit unconventionally, but rather than having to load each of my recordings in neon player prior to analysis, I wrote a python script that will open up the .raw and .time files from all my exported recordings and spit out csvs for the data I need.

I have everything working for basic eye state and gaze data, but I am having trouble figuring out what the columns correspond to in the "fixations ps1.raw" file. It looks like there's 15 columns, but I am not sure whats what. Can someone with insider knowledge on this let me know the column names?

Good afternoon, Quick question about wireless charging — we’re going to be doing several back-to-back hours of recordings using the Neon glasses, and I wanted to ask whether a wireless charger would work effectively with the companion phone while it's connected to the glasses. Has anyone tested this setup? Also, if you’ve got any recommendations for reliable Qi wireless chargers that have worked well with the phone and Neon during long sessions, I’d really appreciate it.

Hi @user-a578d9 👋 ! Whether wireless charging works depends on the companion device. Both the Samsung S25 and Moto Edge 40 Pro support Qi wireless charging.

Just a heads-up: they don’t attach magnetically, and wireless charging is generally less efficient, with the handicap that it tends to warm up the phone.

While this is not an official recommendation. I’ve been using a previous version of this, it helps keep things cool while charging 😉

Good evening, I ran into an issue when trying to time-align my cloud exports with my xdf data when recording using the LabRecorder. As we were able to find a neon lsl stream that was coming from the mobile device connected to the glasses directly, we did not record using the lsl_relay on a separate laptop (how we did it before the app update). Now, after just recording using the stream coming directly from the neon phone, I am unable to time-sync gaze events using the lsl_relay_time_alignment. Is there another way to time-sync the streams without using the lsl_relay on a separate laptop? I could not find information on this topic in the Pupil Labs Companion LSL Relay documentation. This is the error message: WARNING Session id missing! Skipping incompatible Pupil Companion stream xdf_cloud_time_sync.py:107 [{'acquisition': [{'manufacturer': ['Pupil Labs'], 'model': ['Neon'], 'session_id': []}], 'channels': [{'channel': [{'label': ['Event'], 'format': ['string']}]}]}]
ERROR Could not extract any time alignments for [this is my path here]. No valid Pupil Companion event streams found!

Hi, @user-e0a71c - yes, the LSL Relay app is a solution provided for our older eyetracker (Pupil Invisible), but it worked for Neon as well until we integrated the LSL outlet directly in the companion app, so you're correct to avoid the LSL Relay with Neon.

To sync data between LabRecorder XDF data and a Neon Recording, you just need one or more Events that have been recorded in both. Then you can simply examine the timestamps of the same event recorded in both systems to compute the offset. A quick search of this discord server should turn up a few different conversations on this exact topic

The app should send recording.begin and recording.end events (you'll want to start your LabRecorder session before your Neon recording is started and end your LabRecorder session after the Neon recording is stopped). Additionally, all of the events that you add (e.g., through the realtime api, monitor app, etc) will appear in both the Neon recording and the LSL stream.

I don't personally have experience aligning data using just the data, I'm afraid

Good morning! We are seeing some calibration issues in our setup. Person is sitting in front of a 27" (non-curved) screen with a distance of around 40 centimeters. Calibration routine is executed by displaying fixation points at the corners of the screen, asking person to look at it, and correcting mismatch on the phone (following https://docs.pupil-labs.com/neon/data-collection/offset-correction/).

Is this a problematic distance/setup for good calibration?

Hi @user-e0a71c! May I ask if you can elaborate on what you mean by the gaze data recorded in the .xdf file is different from Cloud gaze?

Dear Miguel, thank you, I'll read whitepaper and discussion that you linked. I'm sure it will help me understand the issue. Meanwhile one of my students is trying a posteriori calibration that accounts for the strong barrel distortion. Since we asked all participants to look at corner points at the beginning of every recording, we might have some luck there.

Hey I had this problem too! Did they give any more info? We discovered that it was our office's network's security measures blocking us. We decided to just use a wireless router and make a separate network which works, but then we can't simultaneously use the internet. Would be great to hear another solution on this

[email removed] I work in Matlab and import

Oh my bad, the response got buried! I guess our approach was the right one. Are there any other recommendations to get around network restrictions if we want internet access still?

Yeah so I think there's some larger security restrictions at play that make this difficult to do in our offices. I'll talk this over with my team and try again. But before I was using the wireless router and had my PC and companion device connected to it so the Monitor would work. However for whatever reason I had the ethernet cable connected to the PC and the wifi wouldn't work. I think its because it can't be on two networks at once, but not sure how to get around it. If we come up with a solution, I will message should it be useful to someone. For now im thinking of a work around by downloading all of our experiment somehow

Hi 🙂 I have a couple of questions, some may be a bit silly but I want to make sure I am doing the correct thing. Thank you in advance. If I download the Timeseries CSV and Scene Video are the video and gaze data undistorted, I mean, are they already corrected for lens distortions? From what I see in the videos, they are not. So, I am following the Undistort Video and Gaze Data tutorial (https://docs.pupil-labs.com/alpha-lab/undistort/). What data should I be using for this (Timeseries CSV and Scene Video or Native Recording Data)? Also I am a bit confused about what timestamps of every world video frame means in the context of the world_timestamps.csv? Could you please elaborate.

What is the easiest way to calculate convergence from the eye state data?

The end goal is to examine convergence differences between novice and experts in a medical/surgical setting.

Hi another question here: in the Neon Player is there a way to adjust the offset at varying times and apply it to only a select portion of the video? In our experiments we will have participants looking at a table and later at a screen. The screen gazepoints seem right on the mark, but on the table when the participants are looking down it is a bit inaccurate and could use the offset. Thanks!

In our experiment they are working with legos on a table and looking down, but I think its dependent on the person how much they move their head. We discovered we could get the undistorted image by using opencv and manipulating the images with the API. We're unsure if this would resolve our issue of requiring an offset for when they're looking down at the table. Can we expect this to more accurately display the gazepoints than the fish-eye view you see from the raw video (meaning show the gaze circle over the correct point on the table as it already does when the wearer is looking up at the screen)? We're also wondering: how are gazepoints calculated? Understanding this will help us to make adjustments via the code. Let me know if you have more questions that I can help clarify. We appreciate your speedy responses!

Hello, one of my students tried this script: https://pupil-labs.github.io/pl-realtime-api/dev/methods/simple/streaming/eye-events/ from the simple API but the .receive_eye_events() method does not exist on the device object. In contrast, it works with the async API version. Does the previous code need to be modified in some way? Thank you very much!

OK thanks, it worked! Maybe it would be useful to add a note on the page suggesting an upgrade! Thanks a lot

Oh sorry, i didn't notice that! Thank you very much! I have another question: when exporting the recordings from the phone and transferring them to the PC to view them in the Neon Player, the 3D data appear empty. Am I doing something wrong in this case as well? Actually, the recordings that have been uploaded to the cloud are fine, but we’ve run out of cloud space, so we transfer them directly to the PC via USB

yep

ok.. Isn't it perhaps necessary for the data to be uploaded to the cloud to be somehow "processed" in order to extract the 3D data? I’m asking to understand how the device works

Hi - I'm trying to figure out the precise timing differences between the source neon phone and the client, in order to eventually estimate what the jitter and offsets we can expect when compared to EEG data. I've been looking at the packets over wireshark and am a bit confused. 1 - eye gaze data should be 77 bytes when gaze and eyestate are included, but the closest packets I can find have an 88 byte payload? 2 - Theres an RSTP packet that described data from /?camera=gaze?audioenable=on. Aren't the eye camera stream under /?camera=eyes ? What is that stream describing? 3 - Similarly, what are the /trackID=0 rtsp streams? 4 - There are a bunch of payloads of 200 bytes that wireshark can only identify as udp or unknown rtcp version. Are there any filters/decode as options that would help me analyze this data better? Thanks!

Also, is the timestamp returned by the API as the time the data was taken just the timestamp on the RTP packet?

Unfortunately, we are not using LSL, but some custom thing. We have our own master clock running a ptp server, and an IRIG clock signal gets sent directly into the EEG stream. I am going to change the android phones NTP server to our clock server through adb. and our client computer is running a high time accuracy linux kernel also using the same clock. I'm trying to figure out if the timestamps in the datastream are when the data was taken/computed, or the time on the NTP stamp when the packet actually got sent to the client computer.

Thanks! I'm a little confused though, because the code for RTSPData and GazeData/EyestateGazeData seems to unpack all the variables from the packet in the realtime API, and then just grab the timestamp from _WallclockRTSPReader? Additionally, receivegazeData lists the number of bytes in the packet, and the number doesn't add up to include the timestamp? for example, just GazeData is said to be 9 bytes = x (4), y(4), and worn (1). The timestamp doesn;t seem to come from the payload in the packet?

Thanks for the clarification! The reason I was turning to looking at the packets is because when I wrote a python script to compare the timestamps on the sensor data to the client ingestion time, and compared that to the estimated offset and roundtrip times provided by the timeestimate function, I get close matches for the time offset seen when comparing ingestion time and sensor times for the gaze data (they are like 10 ms more), but the video data seems to be off by like 80 ms. This made me think there was some parsing delay in my python code (even using the async library), so I wanted to compare the times to when the packets were actually received by the computer. So is the timestamp of the sensor data also embedded in the payload of the RTP packet, or is the time attached to the RTP packet just the time the sensor took the data? I'm really sorry for my confusion

Sure sorry for the confusion. This is mostly for post hoc analysis - aligning the neon data with an EEG or maybe other signals that are also synced via PTP. Ultimately I want to know the offset between the sensor clock and the computer/client clock. I know you provide that through the TimeEstimator class, but I'm essentially trying to self verify it as well. From what I understand, there are a couple of times to consider: the time the sensor took the data, the time the data is sent, and the time the data is received. From what you are saying, the time the sensor took the data is in the payload of the RTP packet, separate from the timestamp on the RTP packet (which is when the data was sent?). The time that the computer received the data should optimally be half the roundtrip time, which in my current setup should be < .5 ms according to your API. But there is probably some latency/processing which happens (ideally a smaller amount in the time wireshark received the packet compared to the python API). That latency (based on the time that wireshark received the packets, and the time on the RTP packet, and assuming a clock of 90000 which is in the RTSP description) is currently 2 - 5 ms. I'd like to be able to compare the time the sensor actually took the data to these other timestamps. But I don;'t know how to parse the RTP payloads. Through wireshark I have payloads that are 89 bytes, which matches up with the description of EyestateEyelidGazeData. But that doesn't include a timestamp in it? So is the timestamp on the RTP packet the time the sensor took the data? I.e. there is no difference between the time the sensor took the data, and the time on the RTP packet (not the payload of the packet)?

Sure sorry for the confusion. This is

Thanks for this more in depth answer. Perhaps others in the community have more to share about dealing with gazepoint accuracy in the periphery? Because our participants are looking down at the table for portions of the experiment we’d love to know how we can better handle the gaze offset.

Hi Team, I am not sure what's going on here. This recording has been uploading for quite a while already. ID: bf342863-0439-4edb-88fb-5edcb73c00e3

Also, it says the cloud storage is full. If I deleted my cloud recordings, is it going to help? Or do I have to purchase additional storage?

Hi Miguel - I have decided to interpolate blinks. I am wondering if you have any suggestions for packages/functions which work reliably with pupil labs / neon data?

Hi Team! Not sure if this topic has been touched on before here but I was wondering if there is a way to perform the same processing that is done in cloud on data that has been downloaded as native/raw data from a phone but has been deleted from the phone and is now only available on a computer? I am noticing that for 3 donwloaded recordings I am missing the data from the cloud but I have set up my analysis pipeline in a way that uses the fixations.csv and world_timestamps.csv files from the timeseries download. The neon player allows me to export a file that is also called fixations.csv, i assume this is the same as the one from the cloud agorithm. Can I somehow also create the world_timestamps.csv with the neon player? I only find a world_timestamps.npy and world_lookup.npy but these do not seem to be on the same scale as the world_timestamps.csv. Thanks!

Hi @user-898b32 , Neon Player's CSV formats match with Pupil Cloud's.

To get the world_timestamps.csv file, make sure to also activate the World Video Exporter plugin.

You can also use pl-neon-recording's CSV export script.

Hello 👋 New here. Not sure if this has been discussed, but can the Neon receive TTL pulses? Currently, trying to synchronise their data with intracranial EEG data, and couldn't find guides online. (P.S. Successfully followed this guide with my experiment: https://github.com/pupil-labs/pl-neon-matlab/blob/main/track-your-experiment-progress-using-events.md) Thank you in advance.

Neon doesn't receive analog TTL signals. But it has a real-time API that can send/receive software events over a local network. That said, often the easiest way to sync Neon with EEG equipment is via our Lab Streaming Layer integration, if your EEG device supports that.

[email removed] , there is the typical

Hi, sorry I am having the following error on the screen: May I know why this error happens and if anyone knows a possible solution? Thank you

I tried unpluging it and plug it in back again all morning, but keeps having this error

Hello, I have a question, as shown in the picture is the frame of neon, the temples and frame are separated, on the model do they have a link structure, or do I need to link myself. Thank you so much.

Hi Team, on the Pupil Cloud are we able to set the number of fixations we want to be plotted on the gaze plot? Initially when we ran it through the google collab workbook we were able to select the number of fixations to be plotted. is that option not available anymore?

Hi, all--Has anyone recorded an individual with infantile nystagmus syndrome (a.k.a. congenital nystagmus) with the Neon? A head-mounted tracker with a wide gaze angle operating range would be really useful but it would be nice to see a sample of a recording before asking my department to get one. Thanks!

Hi, I am working with data collected with Pupil Neon and have computed enrichments using PupilCloud to record fixations on AOI. I noticed that my fixations include cases of duration <100ms. I understand that 100-200ms is the treshold for the actual definition of fixation (from your online documentation). So I was wondering how others have dealt with this and whether a simple filter would suffice to clean my data, or whether I am missing something else in the settings of how raw data is used to extract fixation (if so, should I work in PupilPlayer instead before downloading my data / computing enrichments)?

Hi @user-7413e1 👋 Just a quick note, I think there might be some confusion between Pupil Core and Neon.

For Neon, fixation and saccade data are available via the fixations stream which is now computed in realtime too. You can also find more detail in the fixation's detector whitepaper (PDF) where you can find a table noting that the minimum fixation duration for Neon is 70 ms.

Is this on the fixations.csv from the TimeSeries or somewhere else?

That makes sense, fixations in Marker Mapper require two things:

1. A fixation must occur, and
2. It must be mapped onto a surface.

If the surface isn’t detected in a given scene camera frame, due to motion blur, occlusion, or markers being out of view, the mapping can break. As a result, a fixation might not be mapped at all, or only partially.

The ~30 ms you're seeing is quite close to the scene camera’s sampling rate. This could mean that in the following frame, the surface wasn’t detected or the fixation no longer landed on it, so it broke.

You can still recover the full fixation duration by using the recording ID and fixation ID to look it up in the timeseries data.

Hi team, I'm running into some issues with the companion app and was hoping to get some insight: 1. When I start a recording, the duration shown is the time from button press, but it doesn't match the duration of scene capture. For example, a 9s recording will capture 3s of video. 2. The app becomes unresponsive randomly and will get rid of 1-3 of the previous recordings. I am using the Motorola edge 40 pro and I am trying to record 3sec clips back to back, but the cameras activation has been a various times, so I record 8sec and trim afterward. I would appreciate an help

Hi @user-f43a29 . We need each moment of interest to be a separate recording

Hi everyone! I could use some help with a couple of issues I’m having with my Pupil Neon glasses.

1. Connecting Pupil Neon to desktop software: I’ve used the glasses successfully with Pupil Cloud, but now I want to connect them to the desktop apps (like Pupil Service, Pupil Player, or Pupil Capture). I’m not sure how to do it, and I haven’t found clear instructions. Has anyone here managed to set it up? Any steps or tips would be appreciated!

2. Blink detection not working on Pupil Cloud: Even though calibration works fine, blinks are not being detected or logged on Pupil Cloud. Is there something I’m missing or need to activate?

3. Reference Image Mapper not detecting reference image: I’m also trying to use the Reference Image Mapper to track when a participant looks at a product image, but it doesn’t seem to detect the reference image during playback. Any idea why it might not be working or how to troubleshoot it?

Hello?

Hi Team, I am processing a few sessions of neon data locally (v4.8 player and running offline fixation detection since the data are collected with v2.8 companion app).

I noticed that fixation accuracy is heavily influenced by blinking in many cases. In the linked example, it seems like fixation is assigned based on the instantaneous gaze positions post-blink (note these gaze are marked as "blink" by the blink detector), while the more accurate position (with gaze stable for >500ms) is higher above.

Is there a way to customize the algorithm such that fixation estimation ignores the initial post-blink gazes? Or register these as two separate events?