Hi @user-b55ba6! Thanks for your question. Before I answer, I want to be sure I understand your goal. Can you elaborate more on what you mean by locating an object at a certain angle with respect to the eyeballs? Are you trying to compute, e.g. viewing depth based on the eye state measurements?

Hi, how did you do the normalization for x (gaze x [px]) and y (gaze y [px]) corrdinates of gaze in "gaze.csv"? I need to know the max and min value of these pixels to cut the plan into grids.

I am receiving the Lab Streaming Layer from the Neon companion application. As expected and per the gaze data fields, I am receiving 16 channels (fields) per sample. Can someone confirm for me the order of the fields being output over LSL? I checked the docs (https://docs.pupil-labs.com/neon/data-collection/lab-streaming-layer/) but they appear out of date. I took a stab at the order as follows:

"pixels_x","pixels_y","pupil_diameter_left","eyeball_center_left_x","eyeball_center_left_y","eyeballs_center_left_z","optical_axis_left_x","optical_axis_left_y","optical_axis_left_z","pupil_diameter_right","eyeball_center_right_x","eyeball_center_right_y","eyeball_center_right_z","optical_axis_right_x","optical_axis_right_y","optical_axis_right_z"

Thank you!

Hi, I have four projects (=subjects) where the 3d_eye_states.csv file does not contain any values but the headers. This occurred for all the recordings from these 4 consecutive subjects, but not for other recordings collected before or after. Do you have any idea on how to fix that?

Hello, I was wondering if there was any advice for helping the Neon better recognize the April tags? (Using Neon Player)

The lighting in our workspace is my main issue. Our setup requires lights to be off and the tags aren’t able to be placed on the screen.

Currently I’ve rigged up “light boxes” to backlight the tags, but I’m getting really inconsistent results with that.

Is it possible there is some distance/size ratio that the April tags are more likely not to be readable?

Edit: I should add, we’ve tried leaving the lights on or having lamps on in the room to partially light things. However, these results are inconsistent too as it seems the light is washing out the tags

Custom Python code. I do see that the field information is contained in the inlet's metadata. Thanks. I was able to inspect the XML from the inlet description and confirm the order as correct in my original message.

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?

Can we set the scene camera exposure using pupil_labs.realtime_api?

My understanding is that we have two kinds of gaze angle outputs from Neon: (1) cyclopian gaze angle in 2D coodinate [px] that is scene camera coordinate and (2) binocular 3D gaze unit vectors (optical axes for left and right eyes). I'm trying to convert the first cyclopian gaze data to 3D gaze unit vector. Do you have a conversion from pixel to angular coodinate for the first data? Alternatively, is it possible to generate a cyclopian gaze vector from the second data (is this how Neon calculates cyclopian gaze in pixel coodinate for (1)?)

Hi @user-b55ba6. I feel like I might be better understanding your goal. By the sounds of it, you're trying to transform Neon's eyeball centre measurements to the coordinate system of a third-party camera. Is that correct, and is the second camera you're adding a scene camera or an eye camera?

Your understanding is almost correct. I'll provide some clarifications below: 1. Neon's gaze measurements are given as:
a) 2D coordinates in pixel space
b) The angle of a gaze ray in relation to the scene camera in degrees

You can easily convert the latter (b) to a vector by transforming from spherical to Cartesian coordinates, as demonstrated in this example.

1. Neon's eye state measurements provide optical axis vectors, which point from the eyeball centre to the pupil centre. These measurements are independent of gaze, meaning that it's not really practical to transform eye state measurements into gaze measurements. Although this can technically be done with some calibration routine, I don't think it's relevant for your question.

I hope that makes sense!

Hi team, we are getting this message on our cloud on the workspace login for the owner. I was under the impression that we would have unlimited storage for all of our devices? We have currently two cores and four frames. Two frames were the latest additions.

@user-37a2bd we sent an email to the email associated with the order ID. Please let me know if you have any further questions 🙂

Third-party eye camera

Yes, that's correct. Eye state provides the optical axes, which represent the pose of the eyeballs. This does not necessarily reflect the visual axes.

Let me clarify the offset correction. Neon's gaze measurements are optimised for the population average, meaning that kappa angles are implicitly accounted for. However, some wearers will deviate from the population average in terms of both angle kappa and other physiological factors. For these wearers, the offset can be used to improve the accuracy of Neon's gaze measurements by reducing systematic errors that are consistent in both magnitude and direction across the field of view.

Offset correction can only be applied to Neon's gaze measurements, not to eye state. There is no offset baked into Neon's eye state measurements.

Importantly, you need to consider 'gaze' and 'eye state' as separate and independent data streams provided by NeonNet.

I hope this clears things up. Let me know if you have any further questions!

I’ve just updated Neon Companion App and now on the world view on the App the gaze circle is staying on the same position as I move my head and gaze around but is working as it should be in the recording. Furthermore with the Neon Monitor when the page is loaded you are presented with a warning that the Video Preview May Not Be Available in IOS, also the video frozen and the gaze circle is moving in the opposite direction to the movement of your eyes and head

Hi @user-057596 👋 ! Would you mind trying to clear the cache of the Companion App?

If that doesn’t resolve the issue, please follow up by opening a 🛟 troubleshooting ticket with relevant details, including your Companion Device model and Companion App version.

Hi @user-d407c1 how do you clear the cache again?

Hello, sorry I have an issue when I want to delete a recorded file from the trash area. I want to free some space in pupil cloud, therefore delete some recorded videos, then there were moved to Trash area. and then tried to delete them premanently form trash memory by right click on it and select the delete , it gives me an "Internal Server Error" . what should I do for this matter? thanks in advance.

Hi @user-d54741 ! Could you create a ticket at 🛟 troubleshooting with the recording ID?

Hello everyone, I need just a clarification (maybe its a silly question). By buying the "bundle - I can see clearly now " is it possible to detach the neon module and put it in the frame mount for the pico or quest 3? Because it is not clear to me if buying the frame mount for quest 3, for example, the NEST PCB is integrated ( I suppose yes) and so I can then just take the neon module mounted on another frame and mount on the other frame. Thank you very much in advance!

Hi, I used pupil NEON device, to record about 10 situations. I transfered these recording into gaze overlayed MP4 files using NEON player (using 'Fixation detector' and 'World video exporter').

But, a few files can not convert into gaze overlayed MP4, although I can see gaze overlayed movie at my local PC . I can confirm that 'Fixation detector' worked well, but NEON player app crashed in the process of 'World video exporter'.

Could you check my recording files if I send you these files?

left_eye_cam_x would be the horizontal distance of the left eye as measured from the Neon module's scene camera sensor. right_eye_cam_y would be the vertical distance of the right eye as measured from the scene camera sensor.

More info here: https://docs.pupil-labs.com/neon/data-collection/data-streams/#_3d-eye-states

Good morning! I am working with the Pupil Neon glasses. I was wondering if you have instructions of how to include the Picture in Picture with the video of the eye and blinks for displaying in the pupil lab cloud .

Hi team, I have another question about our recordings. We have one recording in which we noticed that sometimes, one video frame to the next stays completely the same while the fixation moves. We think this should not be the case (e.g., in the background, people are walking and suddenly freeze for two frames before their movements continue as if nothing has happened). We have a screen recording that we could share of the effect.

Hello - Recorded a session and received this message when I tried to add to a project: "1 recording is processing, has errors, or blocked. This recording can not be added to the project." I can view the enture recording on the cloud, it seems OK - anything I can do to be able to add it to a project for enhancements? Thanks in advance!

Hi guys. I was chatting with some of my collaborators, and they were interested in the Neon eye trackers, but apparently have some on-site restrictions on wifi/bluetooth devices, so they'd not be able to use the companion android device with the Neon. Is there any work being done to have a version of the recording software on the PC compatible with the Neon? Or would they be better off using the Core version?

Hi Nadia. My colleagues work in Defense, so there's apparently an issue with bringing in the android device onto their lab site, since it has bluetooth/wifi capabilities. The eye tracker itself is fine, but they would need to use one of their existing computers with it rather than the companion device. Is this possible?

In both the monocular and binocular hdf5 output, there are always three times, device time, logged_time, and time. I know that logged_time should be the Psychopy’s experiment clock. The problem is logged_time always have many duplicate timestamps

This is not a big problem since I can interpolate the timestamps myself. However, when there are duplicate timestamps in adjacent rows, which row corresponds to the actual moment that timestamp occurs, the first appearance or the last?

Hi, @user-bda2e6 - there have been some updates to the PsychoPy plugin recently that deal with clocks/synchronization. Could you try uninstalling and then re-installing the plugin so that you have the latest version? Let me know if there are still uncertainties after that

This question: “when there are duplicate timestamps in adjacent rows, which row corresponds to the actual moment that timestamp occurs, the first appearance or the last?”

Again thank you so much for the help

Hi, I'm facing an issue with empty csv files after using the Reference Image Mapper and haven't found a solution through search here. Here's what I have: a recording (with special events) that lasts 1.27 min, "enrichments" successfully identified fixation points on reference image (visually), but 1) I can't visualize the heatmap (it seems there are no points, so I see only the reference image) 2) all csv files are empty. Where could the problem be?

This question: “when there are duplicate

Hi, I am planning a new experiment using neon pupil labs and I am considering using it outside in the dark. Do the glasses work in the dark? Thank you

Hello, I am using Neon during an experiment where participants interact with a collaborative robot in the construction of components. I need to log an event every time a participant looks at the robot (e.g., "look.robot.start" and "look.robot.stop"). Ideally, I would like to generate this event programmatically via the API. However, I am currently unable to reliably track this event, even on the Cloud.

I have tried creating a Reference Image Mapper of the scene and drawing an AOI on the robot, but I have encountered two main issues: 1) I am unsure whether this approach will work, as my AOI is dynamic rather than static. 2) I cannot verify this approach, since I can only see the AOI on the reference image and not on the video. I have also attempted to attach several AprilTags to the robot and create a Marker Mapper. However, in this case, I would need to create different surfaces for different parts of the robot (since as the robot moves, some AprilTags may disappear while others appear), anyway I do not see a way to create multiple surfaces within a single enrichment.

Would you be able to suggest a reliable solution for my problem? Specifically, I need to: 1) Accurately track when participants look at the cobot and log an event each time this occurs 2) (Ideally) Generate these events in real-time via API Thank you very much in advance for your help!

Hi general question: after we finish recording on the Neon for a study, what would be the best recommended method in terms of sharing the data - including the video, and all metrics collected - to other collaborators who might not have an account for pupil cloud? Any simple and accessible way people have used?

Hi @user-cc6fe4 , NeonNet works in complete darkness, thanks to the IR illuminators in the module. The scene camera's exposure can be adjusted to optimize it for dark conditions. It might be easiest to do a test or two and then check in with us again with how it works out.

Hi @user-a09a75 , do you know the position of the robot and the articulation/pose of it's joints over time?

Hello! Could somebody tell me whether it is possible to export the recorded scene camera video with the gaze overlay (the red circle) from the cloud? Whenever I choose any version of 'Download' the only video I get is the scene camera video without gaze overlay. Thank you! 🙏

Hi @user-b57ada , sure, you want to give the Video Renderer Visualization a try. Let us know if you need more info.

Hi @user-b57ada , sure.

• All recordings that you want to run through Video Renderer should be put in a Project. They can all go in the same Project or you can put them in separate Projects for organization purposes. Whatever is easiest for you.
• Then, open the Project and go to the Visualizations tab. Create a Video Renderer Visualization.
• You can then tweak settings, if you'd like, and click Run. By default, it will automatically process all recordings in the Project.
• Once completed, you can then Download the result.

I'd recommend checking out this video for an overview of the Pupil Cloud workflow and if you have not yet had your free 30-minute Onboarding, then just send an email with the original Order ID to [email removed] We can get you up-and-running and answer any other questions you might have.

Hi @user-a09a75 , ok, that information is helpful. If I understood the original message correctly, you are not necessarily using (i.e., consuming) the gaze data in real-time, but rather, saving it for later analysis? If so, then a potential solution with Tag Aligner comes to mind:

• You can use Reference Image Mapper and provide it with a Scanning Recording of the test environment. The test environment should have stable, static features that can be used to consistently localize the wearer as they interact with the moving robot, similar to the examples shown at that link.
• A Reference Image will be necessary to run Reference Image Mapper, but for the purposes of Tag Aligner, it is not crucial that it be a "perfect" Reference Image.
• Then, with Tag Aligner, you can represent the gaze/pose of the wearer, as well as that of the robot, in the same 3D coordinate system and use a software, like Blender, to project ("raycast") gaze into the scene and intersect it with an animated model of the robot arm. This would be done post-hoc.

Whether it is better to use a Scanning Recording made with or without the robot in the scene would only require a test or two.

You could also use something like Head Pose Tracker, and knowing the scale of your scene, you could project gaze onto a playback of the robot's motion that way. However, this requires hanging AprilTags about the scene.

In either case, I'm not sure if Events would be necessary here, although they can be used, of course. Your analysis pipeline could output a CSV with columns like timestamp, gaze on/off robot, position on robot body, for example.

Hi Pupil Labs, I am purchasing the frame "Every move you make" (ID: 20250311035037)

Is the CAD design available? Or other design specs? Do you have an estimation of the accuracy of the physical tool? Or even (if it possible to export/import it) the rigid body definition from the OptiTrack software?

Thank you,

Agostino

Hello.

I'm parsing data from neon companion + pupil cloud.

I want to obtain the average position of each fixation for each eye separately. To do this, I am analyzing the eye data for each fixation, obtained from 3d_eye_states.csv. This file provides the position of each eye separately (eye) and its unit gaze vector (v).

With this, I can make a projection, for example, at a distance of 1000mm and obtain a distant point in the gaze direction of each eye. It would be something like:

objectPoints = eye + 1000 * v

I then project this point back to screen coordinates using:

cv::projectPoints(objectPoints, rvec, tvec, cameraMatrix, distCoeffs, imagePoints);

For rvec and tvec, I am using null matrices since I understand that the eye positions are already given in camera coordinates, so no translation or rotation is needed.

I do this for all data lines corresponding to the fixation. In the end, I compute the average of the obtained pixel points.

This gives me the average fixation point for each eye, in pixels.

However, these points are not symmetric with respect to the fixation coordinates stored in fixations.csv. I expected them to be symmetric. They are near but not symmetric.

Am I missing something in the projection, or have I misunderstood something?

Additionally, if I do something similar using the azimuth and elevation of the fixation from fixations.csv, calculating a distant point from the camera using polar coordinates and projecting it with cv::projectPoints, I do obtain the coordinates indicated as the fixation center in fixations.csv.

What is happening here? Am I doing something incorrectly?

Thank you

Hello! Is gaze data recorded by Neon saved onto a micro SD card on the companion device?

Hi @user-97997c! We don't really have this information – the marker cluster geometries and their pose with respect to Neon's scene camera will need to be established by the user in motion capture coordinates. That said, could you share more details about your proposed use case with the Every Move You Make frame and OptiTrack?

Hi @user-d407c1 - replying to this old message because it is the best match to my question. I've checked a few recordings and have also found a mismatch between the fixations from the cloud and neon player. One recording had ~30 more fixations from the cloud and another had ~100 more from the cloud. It does not appear to be the case that neon player is just giving a longer fixation that the cloud splits into two or because the cloud video starts sooner than the neon player video. (Note: unlike the og question, all of our fixations are over 70ms.) The timestamps of fixations don’t line up completely and even ones that are “close enough” have different durations and x,y coordinates. Watching the videos with fixations visualized shows that they are meaningful differences. Do you know why these differences exits – and, like the original question asked, which is more reliable? Thanks!

Hello - a question ... Is it possible to view a recording immediately after completion on a connected laptop device, before it is uploaded to the pupil cloud? We'd like to allow a pitcher to view a windup and delivery immediately after the pitch is completed without a wait - thanks

Also, if its not possible to view it immediately - is it possible to send the live stream to another app via the neon.local:8080 port?

good morning, does anybody happen to know if and where I can set the detection threshold for fixations in neon player? thx in advance

Hi 🙂

I’m working with the Pupil Labs Neon glasses and the Pupil Labs Asynchronous API for streaming data. There’s an issue with the framerate that isn’t entirely clear from the API guide. While the gaze data is captured at 200Hz, the scene video (which we use to detect the screen in real time) is only recorded at 30Hz. The problem arises when combining these two streams: the recommended implementation syncs the gaze data to the scene video’s framerate, meaning you lose the high-frequency gaze data.

I’ve been experimenting with a workaround to decouple the two streams (still a work in progress), but I’m wondering if there’s a more elegant solution. Specifically, is it possible, by design, to overlay the gaze data onto the scene video while maintaining the full 200Hz framerate of the gaze data? Of course, I could have missed something in the documentation, please let me know if that is the case. Thank you!

Hi, @user-a7588f - I suppose you're using the receive_matched_scene_video_frame_and_gaze function?

Instead, you can use device.receive_scene_video_frame() and device.receive_gaze_datum() to pull the data separately.

If called with no arguments, these functions will block until the requested data is ready, but you can specify a timeout_seconds argument here to control how long the call will block waiting for data before giving up. Using a value of 0 will cause the function call to return immediately - either with the data (if it was available) or None if it was not.

So something like this would get you 200Hz gaze with 30Hz scene video (depending on factors like the speed of your network, workload of the pc, etc):

... snip ...
gaze = None
scene_frame = None

while True:
    gaze = device.receive_gaze_datum(0) or gaze
    scene_frame = device.receive_scene_video_frame(0) or scene_frame

    if None in (gaze, scene_frame):
        time.sleep(0.5)
        continue

    frame = scene_frame.bgr_pixels
    cv2.circle(
        frame,
        (int(gaze.x), int(gaze.y)),
        radius=80,
        color=(0, 0, 255),
        thickness=15,
    )

    cv2.imshow("Scene camera with gaze overlay", frame)
    if cv2.pollKey() & 0xFF == 27:
        break

Thanks Neil - I'm looking to replay the captured video pretty immediately in slow motion on a screen that is larger than the devices running the companion app. Dealing with athletes that are warmed up and can't wait for a USB transfer, they can wait 20-30 seconds, but not longer before getting back to activity. If like to review there last play with them immediately after completion and before the next one?

hi, I am replying to this because we are also using the "Every move you make" frames

We are doing some recordings with our mocap system (vicon) and we can properly track the markers over time. And I was also wondering if there were a 3d model of this frame that you are able to share. Alternatively, is there a correct way to compute the frontal camera 3d position from to the markers' one (I am basically using a centroid computation for now).

Sorry if the question is redundant, I was not able to find any documentation online on that, I am also happy to share the details of our pipeline if someone would need it.

Thank you in advance for the help! Alberto

Hi @user-a09a75 , I see. Thanks for the clarification.

When you use the Video Renderer, have you clicked the Undistort video option? If so, please note that the values in gaze.csv are provided in the original distorted scene image coordinates. If you used the Undistort video option and prefer to work with undistorted video & gaze, then I would recommend checking the Undistort Video and Gaze Data Alpha Lab guide.

With respect to the timestamps, the most accurate method would be to seek to a given frame index and then lookup the corresponding timestamp value in the world_timestamps.csv file. Please note that OpenCV can have some inaccuracy when seeking videos and it would be recommended to use pyav instead.

If you are open to working with the Native Recording Data from the phone, or at least adapting code, then my colleague, @user-d407c1 , has used the pl-neon-recording library to apply YOLO object detection to the scene camera video feed. Please see the attached code.

With respect to LSL, you can send Neon Events to it in real-time. It would efficient, and probably easier for analysis purposes, to send an Event whenever the wearer starts looking at the object and then another Event when they stop.

You could also just send a handful of Events to the Neon recording timeline & the LSL timeline, and then do the object detection offline. You would only need to find the corresponding Events in the two timelines to post-hoc align them.

Please let us know if that clears things up!

Hello, I am wondering if anyone ran into a problem when trying to permanently delete videos from the drive, they dont delete?

I am also wondering if the scanning recording for the reference image mapper has to be from the same device as the videos or from the same day? Ive been having issues where an identical scanning recording does not create a good match for the model, even if the context, lighting etc. are identical I would also appreciate any tips on using the reference image mapper in general:)

Hi there!

We're hesitating between upgrading our older Pupil Core (with the original eye camera on a single eye) to the newer binocular Pupil Core, or switching to a Neon. Could you please help me clarify a few things ?

1/ Is it fair to say that compared to the Core, the Neon is actually a downgrade in terms of accuracy (reported as 1.3° with offset correction for the Neon, and 0.60° after calibration for the Core)? Is it the case that better resistance to slippage for the Neon somehow offsets the loss in accuracy in actual everyday use?

2/ Is it the case that the "Just act natural" frame would be the most suitable for in-lab studies in terms of generally fitting well to all head shapes? I think I understand the "better safe than sorry" frame fits more tightly and could be less suitable for some participants. What about the "Is this thing on" and "Nothing to see here" options - is there a meaningful difference in fit or in accuracy?

3/ Is it correct that I can buy the Neon with a "Just act natural" frame, add an empty additional "All fun and games" frame for cheap, and easily swap between the two?

4/ Is it correct that you have to use the Neon Companion app on a smartphone and cannot connect the Neon directly to a computer? I'm a bit confused about this. Our current setup with the Core is that we use PsychoPy to display auditory stimuli, and at key moments in the task we have PsychoPy send a notification to Pupil Remote which we then use to synchronize timestamps. How exactly could we make this work if we can't actually connect the Neon to the computer? (Note that participants in this study don't actually look at the computer so we can't even use the world camera to synchronize the two.)

5/ After ordering, what is the approximate delay to receive a Core vs. a Neon ?

Thanks for all your help ! :)

It should not matter. But so that we can provide more concrete feedback, would you be able to invite [email removed] to your workspace so we can take a look?

Hi @user-a4e164! Certainly, I'll respond to your points below: 1. No, not really. It is true that Core can get you very accurate data, but this requires very controlled conditions, e.g., head very stable, 2D calibration pipeline (not robust to slippage), short recordings and repeated calibrations, constant lighting, etc. In practical everyday use, Neon provides equal to or better accuracy than Core and it is very robust to slippage. It's just much easier to obtain high-quality data over the course of testing. It also provides rich datastreams. 2. This really depends on the types of studies. Because Neon is modular, it can be fitted into different frames best suited to a specific use case. The frame closest to Core is called 'Is This Thing On'. It has no peripheral visual field occlusion and works well for many lab studies. However, 'Better Safe Than Sorry' is also an option. Both of these frames are more snug fitting that Just Act Natural. There is no difference in accuracy across frames–the same module is used. 3. Correct! 4. Also correct. We actually have a Neon-PsychoPy integration. Neon connects to PsychoPy over a local network; you can use a tethered connection via a USB hub and Ethernet for low-latency communication. But if you just want to send events to Neon, this is possible via our real-time API. 5. Our target fulfilment is within 10 days + 3 days shipping for both Core and Neon. However, we can provide a more concrete estimate when generating a quote. If you'd like, we can also arrange a demo and Q&A session via video call. Just let me know, and I can send a link!

Hi @user-a09a75 , glad to hear it worked out with the gaze overlay!

All of the datastreams from Neon are timestamped from the same high-precision clock, so all of Neon's data are synchronized by default. The Native Recording Data and Pupil Cloud's Timeseries Data are in principle the same, so the approach is equally valid for both formats. Pupil Cloud re-processes gaze at 200Hz and computes some additional streams

The pl-neon-recording library used in @user-d407c1 s YOLO example is a Python interface to work with the open data saved on the phone. Internally, it uses pyav and interpolates when sampling streams

Regarding the LSL synchronization, when Stream to LSL is enabled and you send an Event, then it is automatically saved in the Neon recording timeline, as well as LSL's. It sounds like you are already doing this

Note that converting to different timezones does not change approaches to synchronization. It can even obscure that process. Timezone conversion changes how the timestamps are visually represented for display purposes. For accuracy, it should be easier to stick with the default formats and save the timezone conversion for a final step

Since your analysis is already done post-hoc, I'd personally stick with doing it like that. I'd assume it is easier than migrating to a real-time implementation

If you want to apply YOLO or similar real-time models to your situation, I'd recommend checking out one of our Support Packages. Then, we can dedicate resources to assist with code & implementation. Briefly, whenever gaze is on an object that YOLO has detected, you can send an Event

Hi @user-a09a75 , the recording.begin and recording.end Events will be found in both the Neon and LSL timelines. If you take the difference between the timestamp of recording.begin in the Neon timeline and the corresponding timestamp for that Event in the LSL timeline, then you have the offset between the clocks of both systems. This offset will let you transform Neon timestamps, such as those in world_timestamps.csv, to LSL timestamps and vice versa.

Converting to CET or taking into account start times is not necessary for this step.

Motion Capture w. Neon

Reference Image Mapper

Hi @user-a4e164! I wouldn't necessarily consider that as good or bad. It's true that a closer fit will inherently keep the headset secure, especially during dynamic movements. However, we offer a head strap for the Just Act Natural model that eliminates slippage regardless. Plus, Neon's gaze measurements are slippage-invariant, meaning you can take the glasses off and on, move them around, etc., without needing to re-calibrate. For wearers with less pronounced nose bridges, a head strap should prevent any risk of the glasses moving outside of the eye capture area.

Hello! I'm using Pupil Neon for the first time, and I'm encountering an issue where there's an offset between where I'm actually looking and where my gaze is being detected. I've performed manual offset correction and inter-eye distance measurement in the app, but there's still some offset remaining in certain areas.

https://www.youtube.com/watch?v=08z9t3BZ7OI The video is a screen capture of the recording from the app. I'm actually looking at the position of the white circle. As you can see in the last part, there's still a significant offset remaining in the left edge area of the screen.

Is there any way to address this issue? Thank you!

Hi! I was wondering if any of you could point me towards where I can find the script to convert raw Pupil Lab data into csv files

Hi @user-e13d09 👋 ! To convert from Native Recording Format to CSV files locally you can use pl-rec-export. Lemme know if you need help with

Regarding I'm testing the same vdo clip with 5 people, is it possible to aggregate gaze point of each subjects in the SAME frame of analysis?

Because I love to see distribution of attention at the certain frame

Plz kindly suggesting us

👓 neon

Hi, It would be really helpful if we could develop with the realtime API without needing physical devices. Are there any mocks or simulations available to support this kind of development?

@user-f43a29 Thank you!

I'm interested in doing something similar to what's shown at https://docs.pupil-labs.com/alpha-lab/gaze-contingency-assistive/. Since our team shares the device, I was wondering if there's a way to continue development even when I don't have access to the device.

Also, this is a separate question, but would it be possible to create such an application using LSL? I checked the LSL documentation (https://docs.pupil-labs.com/neon/data-collection/lab-streaming-layer/#gaze-data-outlet) and it seems the output only includes coordinate data. Without access to the scene camera frames, it wouldn't be possible to use surfaces, right?

Thank you, but we want to move the cursor in real-time, so we're thinking of using the real-time API.

Hello! I have been attempting to run Face Mapper data overnight, but Pupil Cloud has been stuck in the "Processing" stage for over 10 hours. I wanted to check if the website is currently experiencing issues. Is there anything I can do to resolve this and get it working again?

Thank you for your assistance.

I can confirm that I am also experiencing the same issue with pupil cloud as @user-fea107 and @user-edb34b. I have been trying to run a RIM enrichment and it has also been stuck on the processing stage for almost 24h.

Hi @user-a09f5d , @user-fea107 , @user-edb34b 👋!

Cloud is currently under an exceptionally high load, which might result in longer enrichment processing times.

First of all apologies, as you may experience some delays until results are available. The team is actively working to speed things up and prevent this from happening in the future.

We appreciate your patience !

Hi, I am working with AOI gaze data, to detect what percentage of gaze data is not detected by the Pupil Cloud enrichment and neon glasses. Is there a way to calculate that? I have tried looking at the data where gaze position on surface x and y are empty, but it was not the case in any raw data. Is there any better way to detect the gaze coverage percentage by the eye-tracking system?

Hi Rob, thank you for your quick support. I mean for quality control of the research study, I need to calculate the percentage of the time in which the eye-tracking system successfully tracked the eyes. I am more interested in detecting how often the gaze was not recognized at all by the Neon or Pupil Cloud. I could not find this metric in raw data individually, but if I know how I can calculate it through gaze.csv from marker mapper or the general gaze.csv data from Timeseries Data folder, it will be great 🙂

Hi @user-665dc2 , to clarify, are you also looking to determine gaze estimation accuracy (i.e., a data quality metric)? Or, do you want to know when Neon was worn in front of the eyes versus lying on a table?

Is there a way to realize when the system failed to detect gaze due to low light, not detecting the april markers, or other external/internal disturbances? for example.

Hi! I'm looking into timestamps for the eyetracker's world video. It seems like world_timestamps.csv has different length compared to the timestamps obtained from neon_recording.scene.ts (world_timestamps is longer for most of my recordings), upon closer inspection it also seems like their values can differ substantially when refering to the same frame in the video. I was wondering if this is expected or if I'm missing something? Any help is appreciated, thanks in advance!

Yes, I’m using it

Testing the same video clip for 4 people and love to see aggregation on the same timeline, if possible on a PupilCloud.

I don’t think that’s what I want to do.

Imagine that we have 4 people who seeing the same vdo clip 1mi. 30 sec (separately 1 by 1).

Then I love to see the distribution of fixate of all 4 people at a certain time.

Could we do it on a PupilLab?

I meant, on a PupilCloud

@user-f43a29

Hi, we are having an issue with our recording. We are recording two 30-minute sessions. While everything works perfectly fine for some recordings, others (and we could not find an error on our end) do not have any events in the recordings. This is most often but not always the first out of the two 30-minute recordings. Specifically, we only get the events 'recording.start' or 'recording.end' but not any of the synchronization triggers via LSL and also do not get any of these.

Sure, Thanks!

Hi all, for the IMU in the neon, are the X, Y and Z directions what you would typically expect? X is subject's left-right / medial-distal, Y is up and down towards the ceiling or floor / superior-inferior, Z is forward-back / anterior-posterior?

Hey there, was wondering if anyone had undertaken a project that involved gaze-contingent displays/stimuli that respond to the eyetracking data in real time. I know the latency of such a system would depend on one's own network, since the coordinates are transmitted wirelessly, but I was hoping somebody could share what latency they were able to achieve with this type of setup?

also if this is not the best channel to send this question, would much appreciate being redirected, thanks!