Hello Guys! I wanted to ask if there is a way to perform the calibration without using a screen. Can I start/stop the calibration process and get if it is successful through zeromq message?

Hi everybody! I am trying to collect data via the network API and I have a problem with the speed of request processing. It takes around 25ms to request and get surface data, but I would really need something below 16ms. Is there a way to do that? Thanks! I tested the speed with the following code:

subscriber = ctx.socket(zmq.SUB) subscriber.connect(f'tcp://{ip}:{sub_port}') subscriber.subscribe('surfaces.Surface1')

recording_time = 5 messages = [] message_time = [] t0 = time.time() while time.time() < t0 + recording_time: t1 = time.time() topic, payload = subscriber.recv_multipart() message = msgpack.loads(payload) t2 = time.time() messages.append(message) message_time.append(t2 - t1)

Hey, while doing this, could you please check the fps graph in the world window drops at the same time as the response time increases?

Not sure if I understand. You are interested in surface-mapped gaze, correct? For that, you need two components: gaze and surface-locations. If one of them is not available, the surface gaze mapping will not work.

I understand, but in order to surface-map the gaze, the surface needs to be detected first. And this can only happen at scene-camera sampling rate. Capture buffers gaze until the detection is done and afterward maps and publishes the gaze as a batch

An alternative solution would be to subscribe to surfaces and non-surface-mapped gaze and mapping the gaze yourself. You could use the last received surface definition instead of waiting for a new one.

This tutorial shows how to convert surface coordinates to scene coordinates https://github.com/pupil-labs/pupil-tutorials/blob/master/04_visualize_surface_positions_in_world_camera_pixel_space.ipynb You need to do the opposite but the logic is the same.

You would need the img_to_surf_trans matrix that is part of the surface datum to transform scene camera coordinates into surface coordinates.

Hi, is it pls possible to get unprojected circle normal corrected for refraction using the current implementation? Like using Refractionizer before there is any pye3d model?

Thank you and are those corrections model independent? Like given different 3d model fit would the circle_3d normal be the same for the same image?

thinking of it again I'm not sure how that would be possible I just somehow think that given the normal, camera ray and fixed eye radius there might be some refration model that would correct the normal

Thank you and do you have any idea if refraction correction is mostly affected by X,Y coordinates of sphere center? I.e. incorrect depth does not matter that much?

thank you both

@papr I've spent some time on the intrinsic matrix issue today. It's a bit premature to be completely confident in my results, but I thought I would share an early report. First, I've found a flip along both the azimuth and elevation. Second, I've found a compression. Again - these are preliminary results, and I need to dig some more. ...if you want to see the jupyter notebook that I used to create this figure, you're welcome to. I've just given you read access to a private repo, pupil-trial-analsis. The jupyter notebook projectionB.ipynb demonstrates my approach.

I would appreciate if you don't share the package without asking.

I can confirm the left / right flip.

Oh, sorry. let's move this to vr-ar.

Hello all, do you know of any online (real-time) fixation- duration/direction processing plugins/software/scripts/etc. used with either Invisible or Core?

Hi everyone, new in PupilLabs. As far as I know, you record videos with PupilCapture and then you import the recoding to pupilPlayer, and then export the results to obtain the raw data about pupil positions and pupil gaze. Currently, I am able to record videos with matlab, but I don't know how to export the data to get the acutal data, using Matlab. Does any of you tried to obtain data using Matlab? do you have any documentation that I can look at? thank you very much 🙂

Hi @user-709f65 👋. You can subscribe to gaze and pupil data in real-time. Have a look at the Matlab section of the Pupil Helpers repo: https://github.com/pupil-labs/pupil-helpers/tree/master/matlab, and specifically filter_messages.m

Hello, I have a data format question: In a pupil labs core record, what is the data format of world.intrinsics and eye1.intrinsics? How can I open that file in python? Thanks for any hints

@user-9eeaf8

import msgpack

with open("world.intrinsics", "rb") as fp:
    intrinsics = msgpack.unpack(fp)

It took longer than expected, but now I understand those functions! Thank you so much for the help 🙂

I have a different question. I want to track the gaze, and I guess the calibration procedure is needed. However, in my current setup I will not use a world camera. Is it possible to perform a calibration, without the world camera?

I see...it might be more complicated than expected. And is there a way to load a dummy calibration? via Capture or via Matlab?

Yes, sure. We work with an haploscope system, what basically shows a different image to each eye to study depth perception, using two monitors and a set of mirrors. I´ve attached an image.

We want to measure gaze while performing some visual tasks of moving objects, placing PupilLabs cameras behind the first mirrors (which are hot mirrors, that allows IR light to go through)., one for each eye. That's why we don't need scene camera, because each eye will have a different scene (i.e. different monitor). The main issue now is that Pupil Labs needs a calibration to start collecting gaze data. Hope everything is clear enough.

Thanks !!

actually i wanted to use pupil core world camera like a web cam. Basically i want to give that as my input to yolo.

so how can i take the real time time video stream as input ? any idea