Hi so the button for Detect Eye 0 is stuck at on and camera isn’t showing, it shows for a few seconds then disappears. The button is also just stuck and doesn’t toggle. Any troubleshooting thoughts? Thanks

Hi! Is it possible to find the angular position, velocity, and acceleration of the pupils using pupil core? Is it also possible to do pupillometry using pupil core?

Hello! I obtained the coordinates of points A and B as (x1, y1) and (x2, y2) using the Pupil Core eye tracking headset , and then obtained the coordinates of point C, but why are its coordinates not within the range of points A and B? Where A and B are obtained by gazing for a long time and obtaining 2000 coordinates to take the average.

Yes

Because I constructed a rectangular region using the points A and B, and the C I was gazing at was inside this rectangle. But the value of point C with high confidence is not in this range either.

In the process of recording the video, I found out what my problem was. Thanks for your answer!

I used 2000 gaze data and then averaged them to get the coordinates of points A and B. But when I look at the world index of these 2000 points, it is not the same as the world index when I focus on that point in the video. I would like to ask if the pupil data obtained through the API cannot be done in real time.

I recommend using timestamps not video indices to find specific sections in your data. Frame indices can be misleading when comparing data from different sources.

We planning to use heart rate and GSR with the Eye-tracking. Can we sync all these together? We have not got the device delivered yet but just curious

Hi, I cannot get pupil capture to work under mac os x Monterey using sudo. Getting the error ImportError: ('Unable to load OpenGL library', "Failed to load dynlib/dll 'OpenGL'. Most probably this dynlib/dll was not found when the application was frozen.", 'OpenGL', None)

The full command/output is: NIWA-1019500~/Downloads$ sudo /Applications/Pupil\ Capture.app/Contents/MacOS/pupil_capture MainProcess - [INFO] os_utils: Disabled idle sleep. world - [ERROR] launchables.world: Process Capture crashed with trace: Traceback (most recent call last): File "PyInstaller/loader/pyiboot01_bootstrap.py", line 151, in init File "ctypes/init.py", line 348, in init OSError: dlopen(OpenGL, 0x000A): tried: 'OpenGL' (relative path not allowed in hardened program), '/usr/lib/OpenGL' (no such file)

... File "OpenGL/platform/baseplatform.py", line 15, in get File "OpenGL/platform/darwin.py", line 41, in GL ImportError: ('Unable to load OpenGL library', "Failed to load dynlib/dll 'OpenGL'. Most probably this dynlib/dll was not found when the application was frozen.", 'OpenGL', None)

world - [INFO] launchables.world: Process shutting down. MainProcess - [INFO] os_utils: Re-enabled idle sleep.

The above error is on Monterey 12.3.1

Thanks for any help

I installed pupil_v3/5-7

Hi, is the a possibility to disable the 'c' and 'r' hotkeys used by Pupil Capture. I'm running an experiment with letters on the same computer, so every time the letter 'r' is pressed as part of the experiment a new recording starts or ends... 🙂

And when is an application 'focused' exactly? Is it out of focus when I'm minimizing Pupil Capture?

The answer is yes. 🙂 Thanks for your help. Just need to make sure that Pupil Capture is not focused during the experiment.

Thanks

I want to be sure I am doing the right thing.

In order to create the active tracking area, I create a single surface that covers the entire area of the stimulus before data export. I assumed the dimensions would be normalized to X =0,1 and y = 0,1. Then I have to use the gaze-position data on the surface. If that's correct, what's the difference between the y and x_norm and the x and y_scaled?

Thanks....let me get it correctly, the scaled values are the actual locations on the screen in pixels. And I can convert it to Cm.

We still need to convert norm values to values in centimeter. I assume we just need to manually measure the distances between apriltags and then use it for the conversion

Thank you.

Then the normalized height and width will be 1 each right ? Sorry for too many questions

I'm using the Pupil labs core and the capture software (3.5.1) on a Macbook Air running 12.3.1. I start capture in admin mode via terminal, and while the scene camera is working, the camera for eye cannot be found. Any suggestions as to the cause?

Hi, I am sorry to hear that. Could you please start the System Information (utility program that comes with every macOS installation), select USB on the left, and check how many and which Pupil Cam entries are listed in the USB Device Tree view?

hi everyone, i have an issue with the world camera. When i try to adjust it, it disconnects for a time or freezes. In the terminal i see the following:

world - [WARNING] video_capture.uvc_backend: Camera disconnected. Reconnecting… Estimated / selected altsetting bandwith : 904 / 1600. world - [INFO] video_capture.uvc_backend: Found device. Pupil Cam1 ID2. !!!Packets per transfer = 32 frameInterval = 327868 world - [WARNING] video_capture.uvc_backend: Camera disconnected. Reconnecting… Estimated / selected altsetting bandwith : 904 / 1600. world - [INFO] video_capture.uvc_backend: Found device. Pupil Cam1 ID2. !!!Packets per transfer = 32 frameInterval = 327868 world - [WARNING] video_capture.uvc_backend: Camera disconnected. Reconnecting… Estimated / selected altsetting bandwith : 904 / 1600. world - [INFO] video_capture.uvc_backend: Found device. Pupil Cam1 ID2. !!!Packets per transfer = 32 frameInterval = 327868 world - [WARNING] video_capture.uvc_backend: Camera disconnected. Reconnecting… world - [INFO] video_capture.uvc_backend: Found device. Pupil Cam1 ID2. Estimated / selected altsetting bandwith : 904 / 1600. !!!Packets per transfer = 32 frameInterval = 327868 world - [WARNING] video_capture.uvc_backend: Camera disconnected. Reconnecting… world - [INFO] video_capture.uvc_backend: Found device. Pupil Cam1 ID2. Estimated / selected altsetting bandwith : 904 / 1600. !!!Packets per transfer = 32 frameInterval = 327868

I am using windows.

does someone know the cause of this?

(the scene camera)

Which is the scene camera. This indicates a hardware issue as the two eye cameras are not recognized by the mac. Please contact info@pupil-labs.com in this regard.

just tested it again. it started when i tried to adjust the world camera (lense or angle). before that it worked fine. In the past it has sometimes disappeared when i restarted the program, but atm it doesnt disappear at all.

yes sry

alright, thanks for your support

Hello everyone, during the recording the camera says "world: unexpectedly large amount of puppil adata..." unfortunately I have not been able to find out in the manual or in this thread what this warning means and whether it is important to deal with it. If I have just searched wrongly I apologize and would ask you for a link to the manual where all the ET releases are described. Thank you

Hello, New user of discord and pupil core. I was wondering, is there a way to stream live data into MATLAB?

We have some examples on how to stream gaze data to MATLAB: https://github.com/pupil-labs/pupil-helpers/tree/master/matlab

I am curious. May I ask what your use case is that you need the data in real time?

Hello, I get data of gaze position in format [number1, number2] from pupil API. I don't know how to interpret these numbers. Is the first number mean X position and the second number mean Y position? The first number has a range between 0-1, but the second does have from 1 and -2, why?

I subscribed to the topic: gaze.2d..01.norm_pos.

Is there published validation material in regards to the core head pose tracking during processing?

@papr Our group would like to do a pupil dynamics and gaze tracking study where the subjects explore environment outside in a city. During their exploration there we want to record their pupil size and gaze direction and follow what features elicit their attention. Can core set be used for this sort of application? Eg, can the pupil data and video of the scene be collected locally without headset being plugged in a computer?

Does anyone have information on the voltage/wattage/amps for the pupil core eye tracking cameras? Not the model with the world view camera, just the headset with the the two pupil cameras. 5v?

I can not find any documentation showing the power requirements in the technical docs/Discord. Would anyone mind helping me out?

Nevermind found it, thank you

https://discord.com/channels/285728493612957698/285728493612957698/824004931353575475

Hello everyone

The image of the computer monitor looks distorted in Player. Although the surface created by the on-screen apriltags apriltags is a perfect rectangle, it doesn't cover the entire screen because the monitor image is not flat. Would this not pose an issue to data exported, given that part of the screens are not within the normalized space?

The visualized lines just connect the four corners. They do not represent the actual edges of the surface. The software corrects for the distortion behind the scenes.

Hi, I use pupil core and it used to work well. But since a few days, the eyes camera are not detected by PupilCapture (while the world camera still works well). In the device manager, I have only one PupilCam detected (Pupil Cam ID2). I tried to re-install the drivers, but it did not solve the issue. I have the intuition that it should not be an hardware problem because both of the eyes camera stopped working at the same time. Any idea? Thank you for any help!

Please contact info@pupil-labs.com in this regard

Hello!

I managed to stream pupil core data at around 100Hz and eeg data at 5000Hz to the LSL lab recorder. I also streamed EEG data at 5000Hz to the pupil capture LSL recorder but got unstable results. What I'd like to try next is streaming pupil-core data to openvibe via LSL. I see that openvibe can read lsl streams. Does anyone have experience streaming pupil-core data to openvibe?

Thanks 🙂

Hey @user-df0510 you should be able to use Pupil Capture LSL Relay to send gaze data from Capture to LSL (and from there to openvibe).

Hello, I have an output for pupil0 diameter normal, but for pupil1 is not. Confidence for both outputs is in the range of 0.8 to 1.0. Output is in the attached images.

Hai, May i know how do we know whether the participant is gazing at the area of interest. From the picture I attached it stated no fixation available. Any advice on this? Thanks

Hi 🙂 To detect fixations, you need gaze data. You get gaze either by calibrating in Pupil Capture or using the post-hoc calibration feature in Pupil Player. Do you use Pupil Capture in your setup?

The circular marker are the calibration targets, yes. Calibration is a step that is independent of surface tracking. Only the latter requires square markers.

You can have the markers attached/being displayed during the calibration (as long as they do not occlude circular ones). My recommendation for the procedure in Capture would be: 1. Setup square markers, setup surface 2. Start a recording 3. Start a calibration (displays circular markers) 4. Finish calibration, continue displaying square markers 5. Perform experiment 6. Stop recording

Hello, I recently started getting these error messages after updating to macOS monterray (not sure if related). Any ideas what's going wrong?

For step 3, do we need to enter calibration mode because it will lead to screen calibration instead?

Thanks, just tried it but got the same error...

sudo open Pupil\ Capture.app

I plotted 2d pupil diameter, and I read data from exported CSV file.

Hi! We're in the middle of an experiment and our Pupil Core have suddenly stopped working! We have tried on several computers. Same error message.

Do you have any troubleshooting tips?

No, both MacOS and Windows.

Worked yesterday...

Ok, we'll try

Now the eye cameras worked, but not the world camera. Any suggestions?

The world camera suddenly connects, but freezez and dissappears. It dissconnects and reconnets.

I’m running into ioHub startup issues when trying to connect pupil labs with psychopy on a macOS system. Has anyone faced the same, could you please share how you fixed it?

Hai, thank you for this. We tried this method yesterday and it worked. However we tried using the same calibration marker as yesterday the software could not detect the calibration marker. Perhaps the ones we used are already folded. Wanted to know, Does the printing quality of the calibration marker matters? Our calibration could not be detected.

PsychoPy Integration

Dear Pupil Labs Team,

no matter which Calibration procedure I use, I don't get my Angular Accuracy higher than 3 degrees.

I planned to use the pupil core to analyse gaze behaviour of radiologists when reporting on MRI images (where they have to identify certain ROIs in MRI images, using two screens).

If I just reach 3 degrees angular accuracy, with a monitor distance of approx. 1 meter from the radiologist, that would mean that the ROI would have to be at least 5 cm large/wide to be detected sufficiently, right? If I am not mistaken, the pupil core should be able to reach accuracies of up to 1° of visual angle, allowing me to detect smaller ROIs.

I followed the instructions on your documentation page and adjusted the pupil detection parameters as best as I could. I also tried all the calibration procedures, apart from the natural features one. What else could I do? Advice would be much appreciated!

I have a dummy question about camera intrinsics estimation, after running this camera intrinsics estimation and clicking on show undistorted images it works great. But my question is, the recorded video does not show the undistorted image, can the newly estimated intrinsic file be imported directly into the recording or should it be imported manually somewhere?

Because our screens do not seem to be strictly parallel in the camera view, in this case when we apply marker to the screen for surface tracking purposes, what is recorded is not a strict rectangle. So I can replace the world.intrinsics file manually and the recording will show up as the preview situation, right?

ok. Therefore, this is basically unnecessary and will probably just make the video look nicer. 😀

Hello! This it is urgent. I'm doing an experiment with people for my Master Thesis but suddenly one of the two cameras does not work. Last time I used the eye tracker was on monday and it worked correctly, but now I just turned it on and this happened

Hello! We are looking into purchasing the Pupil Core Headset, but we have a technical question - apologies if this is not the best place to ask! The app we are intended to conduct research on is not publicly accessible. It is only accessible via machines authorised on our VPN. We currently use ZeroTier for this which simulates a local network. Will this cause any issues when recording sessions (The machines we record on will have access to this VPN) as other solutions we have tried have required the app to be publicly accessible?

I followed the steps, but the same happens

What do you see in imaging devices - and show hidden devices in view options

Do you have access to another laptop (e.g. macos or linux) to plug Pupil Core into? Windows can be a bit picky with drivers sometimes.

I'm going to try to restart the computer

Nothing, the same

This appears in the window when running as administrator

@user-f76ddf do you have access to another computer (macos/linux) that you could quickly connect Pupil Capture to?

Right now I can't

the computer I'm working with is windows, and I use the eye tracker with unity, but all in windows

@papr can you follow up here please?

Yes, that's the case, the camera is not listed in any of those three

I tried with other headset also equipped with the eye tracker and it worked, so definately the problem is the camera

Hi @papr @wrp , I'm currently having a very similar problem (albeit with different errors) with the Pupil Labs VR/AR Eye-tracking for HTC Vive Pro (I posted a detailed description of the issue in the "vr-ar" channel). Would you be able to take a look at it? I do apologise for cross-posting, but it's quite urgent since I'm currently in the middle of testing for an experiment.

Hi! I would like to ask a question about time matching. I currently use two pupil lab cameras to capture eye images (already synchronized) without using world camera, but their timestamps lengths have a big difference (9705 frames vs 8486 frames). May I ask how can I match timing of thoes two videos offline? I also tried to show those two videos in Pupil Player and found the eye overlay (3825 frames) matched two eye videos as I expected. Thanks so much in advance!

Hi Pupil Labs team, I just installed the software on my Windows 10 computer. The world view and eye 0 cameras are connected, but eye 1 cannot. Specifically, it reads: video_capture.uvc_backend: Could not connect to device! No images will be supplied. The same issue exists on a Mac computer. Would you please provide troubleshooting advice? Thanks!

Hi. I want to know how the confidence for the individual eyes is computed. I mean the confidence that should be above the threshold if it is to be used in gaze mapping. Where can I find this in the source code?

Hi! I dont know if this is the right chat, but I am going to give it a shot. I want to connect the pupil core device with Psychopy. According to the psychopy documentation I should be able to just connect the device to my computer, and then choose it via drop-down. Unfortunately the pupil core device does not appear in the eye-tracker drop-down menu. Do you have any idea why? I know I should also contact the psychopy guys with this issue, but maybe I am missing something from the pupil core side

Hello

The distortion seen in the player also adds distortion to the gaze location. Our fixation and gaze plots have gazes and fixations that are not in the right location in the stimulus overlay. The scanpath simply is not accurate but looks accurate on the video

@marc @user-53a8c4 @user-92dca7 @papr is there a lower /upper bound on the pupil size estimate in the detection algorithm of the Pupil Core?

Hi, do I understand correctly that on both devices world and eye0 work but eye1 does not? Or do none of them work on macOS?

Please note that on macOS Monterey, you need to start the application with administrator rights in order to access the cameras. Run this command from the terminal to do that:

sudo /Applications/Pupil\ Capture.app/Contents/MacOS/pupil_capture

What do you use as input for your scanpath?

Hi, the eye tracker drop-down in PsychoPy should just say "Pupil Labs". And it should always be selectable, independent of the hardware being connected or not.

When you want to run your experiment, it is important that Pupil Capture is running and the video preview works for all three cameras. @papr

Hmmm well the option does not show up for me. I tried to select different versions of Psychopy but that did not solve it

A reinstall solved it. Thanks!

Hello, I have a question on blink detection (not sure if this is better suited in 👁 core or in 💻 software-dev but anyway...):

When a recording is opened in Pupil Player and exported (with the Blink Detector Plugin enabled), the resulting blinks.csv file contains a neatly arranged list of blink events, each with onset and offset timestamp.

If however reading the data recorded by Pupil Capture in blinks.pldata using the load_pldata_file function from file_methods.py, the resulting object contains a bunch of single events which are labeled either as onsets or offsets, with each actual blink resulting in several onsets and offsets and the total number of onsets not necessarily matching the number of offsets.

Am I correct in assuming that blinks.pldata contains the "raw" threshold detector output whereas some additional logic/processing is applied when exporting blinks.csv? If so, is there a convenient way to apply this without having to open a recording in Pupil Player?

Hi,

blinks.pldata contains the realtime detected blinks. Due to the realtime constraint, Capture generates a blink event as soon as it detects a blink onset and continues emitting blink events until an offset is detected.

blinks.csv contain the post-hoc detected blinks where the full gaze data is available, allowing the algorithm to look "into the future" and finding the offset for each blink.

You can use this script to run the post-hoc blink detection algorithm on the intermediate data format https://gist.github.com/papr/40ba7d99f572bc0fe388a81aa2f87424

Gaze in scene camera or surface coordinates?

have you plotted those gaze samples on the stimulus overlay? Do they fit? Where does he stimulus overlay come from?

But how did you generate it? Is it a crop from the scene video? Or the same stimulus that you displayed to the user?

And you have adjusted the surface in Player/Cloud to match the stimulus size exactly, right?

Then this is the cause for the incorrect mapping, not the distortion.

Like adjust the height and width in surface tracker plugin?

The Apriltags were on-screen pictures. I only need a single surface which is generated automatically when the surface plugin is active. I don't need to move anything. In this case, the whole book is a surface

So the apriltag markers are part of the stimulus overlay?

Nope.... I'm so sorry. The whole stimulus is just like a square or rectangle, (like the book) having many detractors and targets on it. All the detractors and targets are AOIs. So, I am not interested in what happened in a single AOIs but the overall stimulus. The whole stimulus is a surface. This way we will have both the gaze samples for fixations and saccades in one thread and file

Also I want the dimension of the whole stimulus to be (0,1) X (0,1). That's normalized. The reason for having a single surface

The whole screen space is the surface including the white space (around that book for example). This is because the participant might look at the edge of the book. The whole screen space is the stimulus

The surface automatically matches the screen. Maybe I should adjust it to the bazel. Note that the object (monitor) image is distorted

Okay I encountered another issue. I set up the pupil core eye tracker, calibrated it before with psychopy, i.e. set the calibration process in the experiment routine. When I am running the experiment with Psychopy, I get a new folder within the recordings directory (so like in the same way when a regular pupil capture recording had been done). But when I drag the folder to pupil player, there is no data available to analyze. F.e. for the pupi.lcsv file all headers are available but it contains no rows with data. Also the world video is missing. Do I have to specifiy else before running the experiment?

This unfortunately did not change. The video is also 0 seconds long, which seems weird. Do I maybe have to use post-hoc data in this case?

looks like this

This sounds like the recording is started and stopped immediately.

Thank you. Please contact info@pupil-labs.com with this information

@papr Hi! I would like to ask a question about time matching. I currently use two pupil lab cameras to capture eye images (already synchronized) without using world camera, but their timestamps lengths have a big difference (9705 frames vs 8486 frames). May I ask how can I match timing of thoes two videos offline? I also tried to show those two videos in Pupil Player and found the eye overlay (3825 frames) matched two eye videos as I expected. I also checked the timestamps and according to image contents I found out two very close timestamps (<0.01s) from two eye cameras respectively do not match at all, but the image content of two timestamps with a 0.3s difference from two eye cameras respectively match well. Any ideas why this happens? Thanks so much in advance!

Yes, my apologies. I saw your question previously but have not been able to respond. Could you please clarify how you performed your tests? Do you have source code that you could share?

world = osp.join(data_root, "world_timestamps.npy") eye0 = osp.join(data_root, "eye0_timestamps.npy") eye1 = osp.join(data_root, "eye1_timestamps.npy")

world_ts = np.load(world) eye0_ts = np.load(eye0) eye1_ts = np.load(eye1)

def find_closest(target, source): """Find indices of closest target elements for elements in source. - source is assumed to be sorted. Result has same shape as source. Implementation taken from: - https://stackoverflow.com/questions/8914491/finding-the-nearest-value-and-return-the-index-of-array-in-python/8929827#8929827 """ target = np.asarray(target) # fixes https://github.com/pupil-labs/pupil/issues/1439 idx = np.searchsorted(target, source) idx = np.clip(idx, 1, len(target) - 1) left = target[idx - 1] right = target[idx] idx -= source - left < right - source return idx

sync_eye0 = find_closest(eye0_ts, world_ts) sync_eye1 = find_closest(eye1_ts, world_ts)

A question about the new Core, our lab has been using an older version and recently received the newest model, do the older programs need to be updated as well? Is there additional software that needs to be handled? Or will the data be outputted in the exact same format?

Hi, i need to know if i can use pupil lab open source software with Gaze Point GP3 hardware? thank you so much

Hi @papr , is it possible to get a standalone post-hoc fixation detection python script? similar to this https://gist.github.com/papr/40ba7d99f572bc0fe388a81aa2f87424#file-extract_blinks-py

Hi,all! I have run pupil video backend on Raspberry Pi 4B, I want to send Pupil Core's videos to Pupil Capture on computer through Pi. But I found that my frame rate can only go up to 30fps, although on the computer my core's frame rate can be up to 120fps. How can I increase the transmission frame rate of the video backend, is it related to the CPU performance of the Raspberry Pi? @papr

Hi, thank you for sharing the details! So to clarify: The shared time matching code works for you in the sense that you get very close timestamps, but the issue is that the resulting images are not in sync, correct?

My guess is that your eye video frame identification code (the function that returns the eye video frame for a given timestamp) is not working as expected.

@user-c8166b

Regarding the eye video frame identification code, since the eye video (eye0.mp4) and the corresponding timestamps (eye0_timestamps.npy) have the same length, I assume they match one-to-one. Correct

So you match frames based on indices? What tool do you use for decoding?

I currently use eye0 = np.load(osp.join(data_root, "eye0_timestamps.npy")) to read .npy, opencv cap1 = cv2.VideoCapture('eye0.mp4') to read .mp4, and cap0.set(cv2.CAP_PROP_POS_FRAMES, image_index); ret, img0 = cap0.read() to get one specific frame. Then I assume the specific frame (e.g. image_index = 0) matches the timestamp eye0[0].

Please contact info@pupil-labs.com in this regard

Thanks!

Hi @papr , thanks so much for your previous help! I have another quick question about 2d pupil detection. The output debug image looks good to me (I can see the clear pupil highlighted in dark blue), but the confidence is 0.0 and the returned pupil is a top left point. Any ideas of why this happened? I checked the pupil size is within the range I set (40,80). Thanks a lot!

Hello everybody. I collected some data today using the Pupil Mobile app (which I learned will no longer be developed). I used the same smartphone and I now have four data sets, three of which I can import into Pupil Player (Win 10). However, one import results in the error "[ERROR] libav.mp2: Header missing". I found some ideas on github and tried to delete the folder "pupil_player_settings" and to use an older version of Pupil Player (3.4.). But as you might guess, both approaches did not work - I appreciate any ideas on this issue so that I can import and analyze the data.

In the gaze_positions export csv, there are normalized positions for x and y. Do any of the export fields represent the gaze positions of each eye independently (assuming we were able to record binocularly)? It seemed like the normalized x and y positions represented the joint position of the two eyes together. Also, how does the headset determine the z position of the 3d gaze point in the gaze_point_3d_z field?

hi! in Pupil Player, is there a way to set trim marks by inputting a specific frame range instead of using the seekbar sliders? I can't seem to get the range to be as precise as I want by dragging

Hello everyone. I have a question about frequency amount. For an experiment, I record between 3 and 5 minutes of video to extract pupil diameter in order to analyse pupil dilatation. The problem is that on 30s i have already 12000 data lines and I would like to reduce the frequency in order to analyse longer recordings without getting thousands and thousands data lines. I don't know if it's possible. Thank you a lot and have a great day

Hello everyone. I am a new user to pupil labs. I am trying to use its detector codes to process my eye videos and generate 3d models of the pupils at each frame. Here is my code: https://colab.research.google.com/drive/10A9FLAARcV4zzwFZcN7kCl3uldSSaZxv?usp=sharing

Everything looks fine, except that the radius of my pupil model (a 3d circle) converges to 0.6 mm, which I guess is unreasonably small for a human pupil. Here is my video (eye1): https://drive.google.com/file/d/15SDR_M19eSlybe4Chs_5CoMshwRcRpNH/view?usp=sharing and here is the results I get by processing the eye1 video: https://drive.google.com/file/d/1-1YS-HgyE2QGGI2o7xXaP4fAV53PDlE9/view?usp=sharing

weirdly, when I record the video of my another eye (eye0), I accidently make the video upside down, and the radius converges to 1.5 mm, which I think is correct. However, after I tried to correct this by flipping each frame with the flipud function, I encountered the same problem: the radius converges to 0.6 mm too now. Here is my eye0 video: https://drive.google.com/file/d/1Z1PV5T3dVIH8Xg_zNDu9Y0GF3Wqzxp1x/view?usp=sharing Here is its results (for the original upside down eye0 video): https://drive.google.com/file/d/1fHk62UO3Hd-Lyqco7WY_-TTUmJvGgGjO/view?usp=sharing

I thought this could be because I used the default camera intrinsics, but after changing the intrinsics to my camera's (you can see this in my code) I didn't improve or change the results at all. Could someone please take a look and tell me how to fix the problem? Any help will be greatly appreciated!

Hi there ! I'm really hoping I'm not asking something that's already been asked but as far as CTRL F is concerned, it doesn't seem like it. I'm running Pupil Capture on Windows and not problem at all, it works super fine, everything is cool and easy. But for experimental reasons, we need to switch to Linux to run other scripts in parallel. So far it worked perfectly, until today... Black screen... Nothing written. The window opens when the execution is launched, black screen and then nothing. We tried running it as administrator through the terminal and saw this error: world - [INFO] camera_intrinsics_estimation: Click show undistortion to verify camera intrinsics calibration. world - [INFO] camera_intrinsics_estimation: Hint: Straight lines in the real world should be straigt in the image. world - [ERROR] launchables.world: Process Capture crashed with trace: Traceback (most recent call last): File "launchables/world.py", line 675, in world File "plugin.py", line 409, in init File "plugin.py", line 441, in add File "camera_intrinsics_estimation.py", line 119, in init_ui NameError: name 'idx' is not defined

So, from that we tried EVERYTHING: running sudo pupil_capture, de-installing the app, restarting the computer, re-installing Pupil Capture, updating the whole operating system, updating the installer software, updating python to the latest version (as .py files are concerned) and so on... So, I don't know if you have any idea at all or if anyone encountered such issue. I'm sure it's something very stupid because it stopped working overnight basically... I must be missing out on a very small detail.

Thanks for your answer, if you have one... any other question on this forum including any words of this error message are rergarding another issue.

Oh and by the way, Pupil Player works perfectly fine with the recordings we did before Pupil Capture died 😦

Okay, that's weird because I don't recall running the camera intrinsics estimation but I did actually plugged off a secondary monitor so you must be totally right ! I might have run the camera intrinsics a long time ago when I got an hold on how this whole system worked !

And it actually worked, thank you very much !!

Hi there! I'm working with pupil core data and am having trouble loading one of our larger files (8.33Gb about 65-70min of recording) into the pupil player. On my Mac it just crashes on a pc it try's to load before going non-responsive. Any help/guidance on how to get out the basic outputs would be very helpful! Cheers in advance 🙂

Hello there, we are trying to compare the performance of Pupil Core with a static eye tracker, to see if we can replace the static eye tracker with pupil core. We have three example test cases in mind.

1. For the first test case, we want to compare the performance based on a screen. We use Apriltag markers to define a surface in the camera FOV and for the calibration and validation we use the routine provided in Pupil Capture. We got some good results in this test. Here my question is, the fixation coordinates, that we get after describing the surface, has the origin on bottom left corner of the defined surface, right?

2. For the second test case, we make a use of a plane, the plane is at the distance of 1.5 m from the test person and the camera. So here, for the calibration, we use the world marker. In the docs, it is mentioned that the spiral head motion is best for calibration. We are not quite sure, what is meant by it? Should the markers be placed on different positions so that sipiral motions is the result or should the marker be viewed on a fixed position by moving the head? When we use fisheye camera, the FOV is distorted. Is this accounted for , while calculating the coordinates for fixations? Also, what happens when the test subject looks away from the defined surface and then again back on the surface. Is this a problem ? Should the experiment ideally be done in a way, that the plane is always visible for the test subjec?

3. Can we get the surface normalized gaze positions using the live data streams?

4. Regarding the calibration procedure in the second experiment, the gaze points as shown in pupil capture does not coresspond with the postions of markers on the plane. Is it because old calibration is used and only after correct calibration, one can expect this to happen ?

we define a plane at the distance of 1.5 m from the test subject. We have some fixed coordinates on the plane. The plane is also defined as a surface using the markers. Then we carry out the calibration. For calibration, we put the markers in 5 different positions, where our test subject looks at the calibration points. (we get accuracy of about 3 degrees, I am unsure how this is calculated, could you comment on this too?) For the validation, then we put ask the subject to look at 3 more known point on the planes. The discrepancy ist however very big. We are unsure what is the cause of this. Is there a better routine to perform a test like this?

I also have another quick question.

Could you comment on the following test setup, where the test subject looks at three different monitors, each with some degree of rotation along the Z-axis?

1. Calibrate the camera using world marker
2. define three different surfaces using Apriltags
3. For each surface, the test person looks at a particular screen then read out the coordinates, and compare it with the ground truth

I will send you the recording right away through e-mail

yes. That is what i meant. Thank you for your quick reply.

Hey. I am trying to figure out how the calibration works for the core. I have looked at the "notify" file, and can see that there is "binocular_model", "left_model", and "right_model" with a lot of parameters. How does the calibration work with these parameters?

What I ultimately want to do is reverse the calibration, to get “uncalibrated” data, or maybe be able to transform the data to a generic calibration. Is there any detailed info somewhere about how the calculations are done for getting the calibrated gaze data?

Also more generally, how is the eye modelled? I can see that there is 3D and 2D based gaze mapping. Am I correct to assume that 3D is a model-based approach? Is there for example any way to get the optical axis?

Thank you in advance.

Hi all, I'm having an issue with my Pupil Core:

Pupil Capture will only show one eye window, and only one eye will measure confidence at a time. I can change the camera to the other eye, but only one will be used for eye tracking at a time. Pupil Service will show both eyes at a time happily; I can't seem to find how to open a second eye window in Pupil Capture.

I have tried reinstalling the Pupil Software, and the drivers for the Pupil Core, to no avail. Any help would be lovely 🙂

Problem solved! I used "Restart with default settings". Thanks to the user who mentioned it in the past :)

Hello. I’m new to this and I might have a stupid question but I’ll try. Everytime I want to see the markers and the heatmap on the defined surfaceI have a triangle in center and only there it appears a spot of red heatmap, on the rest of the screen not and I don’t know what to do about it. Can you help me, i want to see the heatmap on all defined surface, not only one spot

Talking about that triangle and the whole recording is like that

The problem is that even if i stare at my phone screen on another corner for a minute, the red heatmap will still be up there where is the triangle. I think it should be normal to have some red spots there too, am I wrong?

Ok, thank you very much!

Hey! We are trying to use our Pupil Core on a M1 Macbook Pro, we managed to get the video through but the following error keeps coming up when attempting to calibrate pupil capture: gaze_mapping.gazer_base: Not sufficient pupil data available. Unsure as to what is being done wrong? Has anybody experienced the same issue? Thanks!

Hi;

I would like to calculate the heatmap for a given time interval in a recording. Hence, I need to somehow find out either how to trim for that specific time interval or a technique that helps out calculating the heatmap. Can you please provide any solutions regarding this problem?

Hey all, I’m conducting a study and want to analyse the number of fixations and their durations on a defined surface afterwards. After exporting the first participants I found two interesting data sheets “fixations_on_surface_Infotainment.csv” and “gaze_positions_on_surface_Infotainment.csv”. 1. What exactly is the difference between them? 2. Which one of them is better for my purpose? I believe it’s the fixations…csv, or am I wrong? 3. Is there already a python or R script or sth. similar to extract the number and duration of fixations on a surface? Thank you for your help! Kind regards, Lena

Hi;

I would like to calculate the heatmap for a given time interval in a recording. Hence, I need to somehow find out either how to trim for that specific time interval or a technique that helps out calculating the heatmap. Can you please provide any solutions regarding this problem?

The relevant operations have been suggested for Pupil Player v3.4.0. Different requirements can be seen in your version. However, I think you will adapt it in one sitting.

Thank you so much for your helping

Hey all, So we are setting up a study where we would like to use Pupil Invisible in our Unity application. Is there a Unity SDK for using the Invisible product line or is that online for the VR product?

Thanks!

Hey! We are setting up a study exploring real-time adaption of gameplay elements based on psychophysiology. As such, we would love real-time gaze data from our Invisible glasses in Unity. We have experience with gaze mapping from remote eye trackers to the images we are exposing to in the experience. Since all our other data is coming in through Unity, we would also prefer that our eye tracking data was too. So we dont have to rely on a connection to a python server etc

Yes, for faster detection, the algorithm assumes a maximum duration. It is configurable via the ui.

Hi all. We are conducting a study in which we want to sync pupil core glasses with other events that are being called in an experimental script that is run in Vizard (python based). I have looked at the Pupil documentation for syncing between Pupil and other software, and they all seem to start with the need to import zmq. We are running our experiment on Windows and I'm finding it hard to import zmq. Any advice?