Thank you very much!!!♥️

Hello,

We're having a few issue with our Pupil Core glasses. When we initially received the glasses, all of the cameras were working correctly. After a few uses, the world camera stopped working. On Windows 11, we consistently get the following message in the command prompt: "world - [ERROR] video_capture.uvc_backend: Could not connect to device! No images will be supplied.". On Windows 11 and Ubuntu 22.04, the eye cameras seem to work, although it can be finnicky and sometimes requires unplugging/plugging in the glasses a few times. However, we cannot seem to get the world camera to work. There was a recent occurrence where the world camera worked, but then one of the eye cameras stopped working. Shortly after that, the world camera stopped working again, but both eye cameras were still functional.

Here are some of the fixes that we've tried: - Uninstalled and reinstalled the latest version of the Pupil Labs suite - Reset the Pupil Capture settings to default - Deleted the pupil_capture_settings, pupil_player_settings and pupil_service_settings directories from the home directories. - Uninstalled and reinstalled the "libusbK USB Devices" drivers - Disabled and re-enabled the cameras via Device Manager - Used the glasses and Pupil Capture on different hardware/machines and operating systems (e.g., two desktops running Windows 11, a desktop running Ubuntu 22.04, two laptops running Windows 11). - Tried the USB-C connection as well as the USB 3.0 adapter - Tried a different USB cable - Tried different USB ports - Reconnected the eye cameras and inspected the pins. - Searched through the troubleshooting section of the Pupil Labs documentation and some of the GitHub issues

All of our machines' drivers are up-to-date. Please let us know if there are any other ideas or suggestions that we should try.

Thanks,

Brayden

Hi, dear pupil lab teams. I'm wonder if you sell the camera world replacement. I think that our camera CMOS was crashed... 😔

Hi @user-cb8fad! Sounds like you've already done some thorough testing. We might need to receive the system back for inspection/repair. Can you please send your original order ID to info@pupil-labs.com and someone will assist you from there!

Yes we can replace world cameras on the core headset. Please reach out to sales@pupil-labs.com mentioning that you require a new world camera!

Hello, how can I generate heatmaps from the eye tracking recorded video?

Hi @user-757948! You'll want to check out the Surface Tracker Plugin. More in this section of the docs

Hi there, the confidence level of the right eye is really low. In the attached video, you can see that even if the camera is pointing exactly at the eye, the model keeps failing to track the pupil and the eyeball. Do you have any idea how to solve this issue?

Recommend setting the eye camera to autoexposure in the settings - the image looks somewhat over-exposed.

Hello team, is there a way to mark areas of interest without using the apriltag marker? I have a puzzle, and I want to analyze every piece. Thank you!

Hello, We are running 2 experiments using Pupil Core: Experiment 1 is performed on a screen with gray background, whereas experiment 2 has to be performed on black background, both in almost a pitch dark experiment cabin (except the light from the screen). 3d pupil detection works significantly better in Experiment 1 where the pupils are smaller due to the light coming from the screen. In experiment 2 with black background, pupil detection confidences drop below ~0.7-0.8 quite often, which leads to time windows during which gaze detection does not work (which we also use for our experiments). In this case, which parameters and settings would help me the most to improve pupil detection in a dark environment?

In the screen capture you shared, it looks like the eye cameras have different exposure times. The one on the bottom appears over-exposed. Are you sure both were set to autoexposure?

Hi! I have a simple question about the right-hand Player (v 3.5.7) menu. It seems to be only partially visible, so I can't access everything I need to (e.g., the manual edit mode toggle). I've adjusted the window size, restarted Player, restarted my machine (MacBook Pro 2021, v 12.6), and toggled different menu options. Is there an easy way to fix this?

You can resize that panel by dragging the three bars to the desired location (see red arrow in screenshot)

Thank you. If I have already passed some participants and have not used those marks, is there a way to delimit the areas of interest?

The camera positioning looks really good, and the exposure also seems to be appropriate. The next steps would be to try the following: 1. Tweak the 2D pupil detector settings - I'd focus on making sure the maximum expected pupil size is appropriate. You can find further instructions/details in this section of the docs. 2. Set the ROI to only include the eye region, excluding the dark corners of the image. Please note that it's important not to set it too small (watch to the end)!

Hi Does it affect the use of an eye tracker if one eye camera is broken and only one eye camera can function properly? Thanks

Hi! I have a question about mapping dynamic areas of interest with data from Pupil Core. I've collected data from children and mothers wearing Cores with apriltags attached at the top of the headsets. We want to see if we can measure looks to each other's faces - the apriltag marker provides a rough estimate of 'face in worldview', but this is a bit less precise than we'd ideally want. Do you have any suggestions about methods for defining areas of interest in the worldview that we can then map fixation data onto? If I'm not mistaken, Cloud is only compatible with Neon. I'm looking for something like this but usable with Core. Thanks so much!

Hi, I am interested in how the gaze estimation is calculated from the pupil position data. More specifically, how is the frequency of both related. Lets say the pupil position data is recorded at approximately 120hz, what will be the frequency of the calculated gaze position data?

Hi team, i would like to know how many seconds/miliseconds of the beginning of the recorded pupil data can be used as a baseline value. Also is it best to use a mean value. The goal is so that i can then calculate the percentage of pupil diameter change compared to the baseline. I did not record a particular baseline value but there was at least a 10 second pause before the next scenario was played to them. Thanks

Thank you. I'm uncertain if I've grasped it correctly, but it seems I need to place markers, such as one in each corner of the board, and delineate each of the pieces individually. Can I complete this task later, or does it need to be done before the participant proceeds with our experiment? Additionally, should I perform this task with each participant or for each recording, or is there a way to generalize it? Apologies for any inconvenience. Thank you.

Thanks btw!

I encountered a strange problem while stream data using the LSL plugin. When I look at the xdf file, the timestamps are not monotonously increasing. Did anyone encounter such a problem?

Hi, @user-6d3681 - if I recall correctly, LabRecorder doesn't make any attempt to sort data as it arrives. This task must be done later.

If you think about it, this is preferable. Because the data is sent over the network, the order of their arrival is not guaranteed to match the order in which they are sent. LSL and LabRecorder are designed to work with multiple high- and low-frequency streams. Sorting data as it arrives can be computationally expensive, and could always be done later instead. Because of this, LabRecorder instead uses its resources to just ensure that all the data is recorded

Can you share or link this load_xdf.m? If that function is meant to sort and is returning data which isn't sorted, the error must be there, right?

load_xdf.m is here: https://github.com/xdf-modules/xdf-Matlab. It has been used by many others and not just me. As I mentioned, it works well for my other streams. I think, also, that LSL does ensure samples are recorded in order and handles retransmits as needed. Does the Pupil Labs LSL plugin not transmit data in order?

I understand that, but, again, all the other streams seem to be fine. They also have more channels and higher data rate.

That may be a bit unexpected, but not impossible. Regardless, if load_xdf.m is supposed to perform a sort, then it wouldn't matter how the data is sent or recorded.

exactly. That's why it's strange. I asked here in case someone noticed that problem in the past. In our end we are experimenting with some other computers (perhaps it's a computer problem). I'll let you know if we resolve it.

manual

hey pupil labs, i just wanted to know what excatly happens when the Core eye tracker dosent recognise the eye in the middle of a recording?

i meant will it record it as a false data or it just skips the perticular point? like will it give a false fixation or blink anything like that or it just skips?

Hi Pupil Labs,

I am not able to find documentation on the Core's working distances. I would like to set a visual task closer than 40cm, ideally 20cm away from the observer's eyes, has this been tested before?

Can anyone please help me out? 🙂

Thank you!! On more question please, when I download the eye tracking recording from the cloud, it gets downloaded without the gaze and fixation points, is there a way I can download it as it is from cloud?

Neon

Thanks for clarifying @user-757948! Let's move this discussion to the 👓 neon channel then. I replied there: https://discord.com/channels/285728493612957698/1047111711230009405/1215222805583499264.

Hi, I would like sync pupil eye tracker with other device, how to do it the right way?

Hi @user-756d06 ! Can you kindly confirm whether you are using Pupil Core or else, such that I can link you to the correct documentation? There are multiple ways to sync with other devices, being the most popular Lab Streaming Layer or iMotions. What other device are you looking to sync with?

I use Pupil core

Hey guys. Could anyone give me a little information?

Hi I was interested in the mechanism of the pupil core device. I wanted to know how does the system decide the confidence value? Is there a computational algorithm which it uses to assess the confidence based on height width and depth of the world view and the pupil position or does it use some form of computer vision based model to provide a reding for the confidence?

Hi! I am interested in achieving the data of pupil and gaze through pupil core by using my own cameras. Is there any demonstrations or examples for testing? And where can I find them? Thanks sincerely!

Thank you very much. I tried using the AprilTags initially, but encountered continuous issues, including poor detection. Is it necessary to define the area beforehand, or can they be analyzed post hoc as long as they appear during the recordings? Thanks!

Hi, Nadia! I am now using Pupil Capture for gaze tracking to show the image in the fovea region through my own AR-NED, and it is critical for me to get the gaze data. From the user guide, I know that I can view and record my real-time gaze and pupil data from Pupil Capture through record function, but I don't know which document has the data I need and how can read it. Could you give me some tips or demonstrations? Here I give you my record outputs and the appearance of my AR-NED.

Hi Peter @user-e91538 have you solved your problem? I think I have a similar issue: the world camera is not working, but the eye camera is. It would be helpful if you could share how you fix your issue. Thank you. Best, Jack

Hi I have updated my Macbook and installed the new version of Pupil capture, now I faced on problem in local USB (it is seem inactive). How can I solve this problem?

Question: How can I normalize the left/right pupil size? In my figure, each green line is a stimuli onset, (this recording has multiple trials) , blue scatter plots are left eye recording, red plots are right eye recording. You can see left and right pupil size are very different, but I'm guessing it's from the camera angle and other instrumentation errors. So my question is what's the best way to match (normalize) the two eye sizes? I'm thinking shifting the red plot baseline up, then rescale the plots so it has the same peak sizes as blue plots, and do it for every single trials. Is there a better way to do this?

Hi Rob, we have a dichoptic stimulus setup, presenting two different images to each eye. The experiment is in a relatively dark room. Each of our trial last 6 seconds, and each experiment has 50 trials, so that's about 5 minutes per recording. The reason for normalizing the two eyes is that we want to be able to compare the eyes' time series directly. The 1mm baseline shift is obviously not real, so we need to fix that in order to do any comparison.

Hello, i'm currently doing my master thesis and im implementing the Pupil Core headset within an entire system of biosensors for monitoring in real time data from First Responder (nurses, firefighters etc...) My objective is to obtain data in real time with the Network Api to MongoDB - Docker. The thing is that im trying to run the Pupil Core in Visual Studio Code and when installing the requirements.txt appears an error. This error is related with the installation of uvc and the wheels. Im lost. Im on windows 11, Ive tried with virtual environment, installing wheel, installing cmake, -m build, the steps you provide in github in the uvc windows section but nothing. i can send you the error image. Could you help me im a bit desperate. Also, could you recommend me how do you think I should implement the system to upload it to Mongo and Docker, should I do it in visual studio, should I do it using the Capture app, I'm a bit lost with that too because I really don't know what to do after running it in python. How could I approach it? Very thank you in advance

Thanks for your quick response. My Macbook is 12.7.2 . I tried troubleshooting but did not work.

Please in which way can we record voice using Pupil Lab Core?

Hello, I’m developing an application that interfaces the pupil capture. The application can start, stop recording, and perform other functionalities. Now I want to develop a new functionality: can close and open each eye camera. Is it possible to do that with the Network API? Thanks in advance, Nobby

Hi @user-cdcab0 , i have a short question regarding the "Pupil Core" model. Is it able to measure the average blinking rates of participants?

Hi @user-7daa32 ! Pupil Core has no microphones, if you would like to capture audio. It is possible to collect audio externally using the Lab Streaming Layer (LSL) framework. This would provide accurate synchronisation between audio and gaze data, but takes more steps to set up:

1. Use the AudioCaptureWin App to record audio and publish it via LSL
2. Publish gaze data during a Pupil Capture recording with our LSL Plugin
3. Record the LSL stream with the Lab Recorder App.
4. Extract timestamps and audio from the .xdf and convert to a listenable format
5. Do post-processing, e.g. make annotations at given sound stimuli etc

Hi @user-afe8dd ! Pupil Core does not output directly the blink rate, but computing this parameter is trivial.

Firstly, I 'd recommend learning about the blink detector employed and how it differs from Capture to Player and the best practices when employing the blink detector.

Once you understand how blinks are detected, and know whether you would like to compute blinks in realtime or post-hoc, you simply need to iterate over the blinks and their timestamps.

Simply define the period you want to compute your blink rate from, and look up for the blinks whose start and end ts are within

@user-d407c1 thank you for your answer. for my research this parameter is quite important. So i guess, i would have to look at the number of blinks a particpant has made, then divide it with their time they spent during the experiment. And do this for all of my 200 participants. So this is quite doable, do i understand correctly?

That sounds like a good approach to get a blink rate per session. Perhaps you would like to compute the blinks per minute across each session, as it might be more useful. That would give you a metric to compare not only across participants but also to compare along the experiment.

Depending on how long the sessions were, how exhausting it was, or whether screen tasks were involved, you may see a decay within the participant alone.

I'm getting the following error when running 1 (out of 21) recordings. Is this a corrupted file issue, and is there a way to fix the file in question?

player - [INFO] launchables.player: System Info: User: ander, Platform: Windows, Machine: Muttnik-3, Release: 10, Version: 10.0.19041 Fixation detection - [INFO] background_helper: Traceback (most recent call last): File "background_helper.py", line 73, in _wrapper File "fixation_detector.py", line 170, in detect_fixations File "fixation_detector.py", line 170, in <listcomp> File "file_methods.py", line 293, in getitem File "file_methods.py", line 253, in _deser File "msgpack_unpacker.pyx", line 195, in msgpack._cmsgpack.unpackb UnicodeDecodeError: 'utf-8' codec can't decode byte 0xbf in position 2: invalid start byte

player - [ERROR] launchables.player: Process Player crashed with trace: Traceback (most recent call last): File "launchables\player.py", line 722, in player File "fixation_detector.py", line 534, in recent_events File "background_helper.py", line 121, in fetch UnicodeDecodeError: 'utf-8' codec can't decode byte 0xbf in position 2: invalid start byte

player - [INFO] launchables.player: Process shutting down.

Hello all, I plan to use Pupil Core when the participants are in motion. I realize that the Pupil Mobile application is no longer available on the Play Store. Is there any way to use this Pupil Core on mobile? Thank you.

Hi @user-75ea17! Do you already have a Pupil Core system or are you scoping out a new purchase?

Hi… I’m going though the GitHub scripts to find out the pccr (pupil center corneal reflections ) implementation… can any one please help me… I’m trying to implement it for a two camera system and hoping to find some help here… would be grateful for any useful leads …. (Note: I am referring to Guestrin and Eizenmann’s paper , but I’m not getting expected results)

Thanks for clarifying. Using Core with participants in motion is not an ideal situation, but is possible. Some of our customers have had success using a small form factor tablet-style PC in a backpack, for example.

No problem. I'd also recommend using the 3D calibration pipeline as it offers better slippage compensation, and breaking your experiment up into smaller parts if possible. You can find more details in the Best Practices section of the docs. Let us know if you need any more help!

Hi, where do I find the serial number of my pupil core device?

Hi @user-ea64b5! You can't find Pupil Core SN by visually inspecting it. May I ask why you need the serial number?

You will need to use the original order ID, which acts as serial number for Pupil Core.

that one yes

Hello! Our lab has the eye tracking glasses from Pupil Labs company. We would like to conduct an experiment using these glasses and other equipment in combination, namely: GRS (galvanic response) and PPG (photoplatysmogram). Which brands of GCR and PPG are the glasses compatible with?

Thank you very much!

hi all, is it possible to use pupil player in a headless fashion to automate data exports. Have a large batch of recordings that I need to get pupil_positions.csv files for, similarly need to do post-hoc gaze calibration (marker dection, calibrations, mappings)

Hello ! I am working with surface tracker and I was wondering if there was any recommendations regarding the tag size in relation to the eyetracker-screen distance Thanks

Hello! I'm not sure if this Discord will be able to help me with this issue, but I thought I would give it a try. My Computer Science Capstone class is working with our Psychology Department to use the Pupil Core glasses. My team split up some work and asked me to find information on the following:

*Documentation on the origin point of all cameras (eye0, eye1, world) and what the correlation between them is. (I think they are asking, if your pupil is at the center of the sensor, are you staring at the center of the screen?)

*The device we use creates a grid in the device itself for tracking the eye movements, we need to figure out how these X and Y values relate to the pixels on the screen the person is looking at. First, I would like to see if you could find any documentation on this, or if there is a script already available to use.

Basically, I am wondering if there is any more indepth documentation then what is on the website?

Any help is appreciated!

Hello! I have a difficult problem when using the GUI of Pupil Capture. Here is the description: I'm now using Pupil Capture for the debug of my world camera and two eye cameras. It is correct to get the image for three cameras in world GUI, but I cannot change the eye camera when it comes to the eye GUI. Actually, neither eye 0 GUI nor eye 1 GUI is working correctly even I change the different USB for each cameras. The appearance is that I can only debug one camera. Has anyone ever encountered this problem? Please give me some suggestions! Thank you sincerely!

P.S. : The drivers of the used cameras have been changed into libsdk. My computer device is ASUS TUF Laptop using AMD 7940H and Nividia RTX4060, the version of Pupil Capture is 3.1.16

Hello,

I’m currently using the camera from Pupil Labs along with the UVC Python library to capture video. It’s working well, and I can successfully capture video streams from both the left and right cameras. However, I’m encountering an issue where the video from the right-side camera is upside down.

For context, I’ve created a Camera class that takes idcam as a parameter, which can be either 0 or 1. The expectation is that the video streams from both the left and right cameras should be identical when I instantiate this class.

I appreciate any assistance you can provide on this matter.

Regards, Nobby

Hi @user-b02f36 👋 ! Could you open a ticket in the 🛟 troubleshooting channel? Then, we can assist you with debugging steps in a private chat.

Hi @user-80123a 👋 ! That is because that camera is actually mounted upside down. You can simply flip the image in software (try cv2.flip(img, 0) or np.flipud() ). However, just to be clear, the video streams from the two cameras will never be identical. May I ask what you are planning to do and why do you prefer PyUVC over Pupil Capture?

Yes, the Pupil Capture software performs the flip already for you. You can change the "flip" setting in the settings of each eye camera window. Is the gaze data provided by the Network API insufficient? Or, is there any reason why the Frame Publisher plugin for the Network API is not appropriate for your purposes? What are you doing with the eye images?

Hi there. I'm just starting to experiment with AOIs and Pupil Core. My experiment setup is quite "natural" (people sitting on a chair gazing at a white wall, whilst thinking about personal memories). I had two questions : 1) I see there are a variety of AprilTags - how does one pick a specific one ? Do you need to use a different type of tag everytime ? 2) how many should I place on my white wall ? Thank you so much for your help !

(Another quick question : does anybody know of this Plugin : https://scholarworks.calstate.edu/downloads/2j62s7153 'Gaze Features Exporter" and know where/how I could install it to directly export average fixation durations, saccades, etc. ? Thank you so much)

Hi @user-b02f36! Just to be clear, you're using third-party cameras with the capture software, not those from Pupil Core?

I'm not familiar with the work you've linked to. What exports exactly does it provide that the stock Pupil Player doesn't?

Thank you very much for your answer @user-4c21e5 ! I believe this work provides data on saccades, which Pupil Player does not ? It also should automatically provide statistics on average fixation duration, average saccade amplitude, average pupil dilation... To avoid doing all the calculus manually. Do you know of any way to do that ?

Hi everyone, I have some data from my last study (on painting and eye tracking), I put my data on pupil player and now I have some questions about extract data. I had 60 images and in extracted data I have fixation_on_surface 1 and fixation_on_surface 2.i can't find what is this. can you please help me?

Thank you so much @user-d407c1 just one more question, I have all the 60 images in one task (aprox. 8 mins). so I must surface 1 and surface 2 just one time and it will be defined for whole task?

The surfaces you define are there for the whole recording duration. Do you have 60 surfaces on an 8-minute task? May I ask how you define these surfaces? Do you use unique markers for each one? Or do you use one set of markers and crop the area?

Hi team. I would like to generate scan paths, however I noticed from this https://docs.pupil-labs.com/alpha-lab/scanpath-rim/ that it uses Reference Image Mapper in Pupil Cloud which is used with Neon and Invisible. I have Pupil Core. Is there a way to generate scan paths with Core?

Hi @user-3c6ee7 - have you checked this tutorial ? It shows how to generate scanpaths with Pupil Core data.

Hey guys, we are currently trying to run pupil_capture on Ubuntu 22.04 but somehow it's not working. First we tried to install the latest release (https://github.com/pupil-labs/pupil/releases/tag/v3.5) which doesn't start due to "Segmentation Fault" in CPython. Then we tried to run the master branch from source: The program starts but we can't open the options of the three different windows (world, eye0, eye1). If you already answered this, sorry for asking. We tried to search this commuity for answers but unsuccessfully. Help is appreciated. Best regards, Rob

Hi @user-1386c3 ! This is probably a different issue, but have you already tried these troubleshooting issues.

hm. It's kinda strange. It doesn't seem to have to do with the cameras. It worked initially when setting up Ubuntu 22.04 (with the release package) but after dist-upgrade it stopped working. Do you run the source on 22.04?

Hi everyone. I have one more question. I extract my data. I have one problem; for pupil _positions, in the clumn of CSV file with the name of method, I have two sources of data: 2d c++ and pye3d 0.3.3 post-hoc, so in the part of dimetr_3d I have empty rows. is it the problem of recording process? can I resolve it in the pupil player or capture?thanks in advanced

Hi! I have a question in Pupil Capture. From developer document, there are two data related to gaze tracking. Due to my use of Pupil Capture for achieving foveated rendering, which gaze data should I use? 'gaze_point_3d' in Gaze Datum Format or 'gaze_on_surfaces' in Surface Datum Format?

Assuming the surface you're tracking is the screen on which you're doing foveated rendering, you'll want to use the surface gaze data

hm. It's kinda strange. It doesn't seem

Ah, so in your case the surface you're tracking is not the display surface then, right? I'm not very familiar with AR-NEDs, but I imagine that you're virtual image has a virtualized real-world position and size then. If so, what you'd probably need to do is calculate the intersection of the 3D gaze ray with this virtual display surface

Yes, although to be abundantly specific, you'll want the real-world location of the virtual image plane relative to your subject's eyes

I'm not sure I fully understand the question. Are you asking if you need to setup a physical screen that matches up with your virtual one? If so, then no, that wouldn't be necessary unless you wanted to validate your math or something.

Hello, we're looking for recommendations for small microphones to attach to the eye-tracker. Does anyone use anything like this?

Hey all, I am working with the pupil core for a class project. I am wondering if there is a way to determine what pixel the user is looking at. I have access to pupil_postions.csv and gaze_positions.csv post recording, and I want to take that data and translate it to determine what pixel (on a screen/monitor) the user was looking at during the recording, or I am wondering if there are any plugins that would help with this? Thanks!

Yes, all of those are indeed in the world camera coordinate system. You got this! 🙂

Hi @user-b02f36 👋 ! It looks like you instead want “eye_centers_3d”, with an “s”, as the key.

I would guess that in the first video, both Pupil Capture and your Python script are running on the same PC, while in the second video you're running Pupil Capture on the first laptop but your Python script on a different laptop?

If so, you will need to change the IP address in the script running on the second laptop so that it uses the IP address of the laptop which is running Pupil Capture.

"localhost" means "connect to this computer", which isn't what you want if Pupil Capture and your Python script are running on separate computers.

I may have been a bit unclear. The fact that you're running that same code on both computers is likely the problem. You can only use "localhost" in that script IF Pupil Capture and the Python script are running on the same computer at the same time.

If Pupil Capture is running on a different computer than the Python script, then you will need to modify the Python script so that instead of "localhost", it uses the IP address of the computer which is running Pupil Capture.

After changing "localhost" to the IP address of the computer running Pupil Capture, you should be able to run that script on any computer on the same network - whether it's the computer running Pupil Capture or not.

I'm terribly sorry to disturb you, dom. I have changed the IP address of my second laptop into the code. Here I gave a breakpoint before the if function I used for separating the data, and obviously I could achieve all message. However, when I removed the breakpoint and made line 31 into comments, I could not achieve the message like the first video shown before.

I believe I see the problem with your code. You are using bytes as your dictionary keys for the message dict, but I can see in your output window that the keys to that dict are actually strings. In other words, b'ellipse' is not the same as 'ellipse'

So on lines 33-34, you have

if b'ellipse' in message:
    ellipse_center = message[b'ellipse'][b'center']

You should change these to:

if 'ellipse' in message:
    ellipse_center = message['ellipse']['center']