Hello. Currently I'm using mac and try to run Pupil from Source, and get this error "Could not build wheels for av". Here is the log file. May I ask how this issue caused? Thanks!
Please use the develop branch. It is also recommended to build a virtual environment with the intel86 python version
Then there are wheels for all dependencies
Thanks for noticing! Currently using Conda for virtual environment
Check out this article on how to setup x86 environments with conda on mac OS https://towardsdatascience.com/python-conda-environments-for-both-arm64-and-x86-64-on-m1-apple-silicon-147b943ffa55
Thank you so much!!
I made it work yesterday! The frame rate seems locked at 30fps
For which camera?
HD-6000
Yes, that is an older model with a maximum framerate of 30 fps
Haven’t removed the case yet. Just a initial test.
Just out of interest, are you building the DIY version?
Oki
Yeah, still looking for another hd-6000 haha. Btw, I also saw some camera module OV9281 online. Maybe it’s perfect for this use case too
Hello, I'm using Pupil Invisible to detect the blinks and I would like to export the frames of the video "world" when the eyes are closed. So I installed Blink Detector (https://github.com/pupil-labs/pupil/blob/master/pupil_src/shared_modules/blink_detection.py) on Pupil Player. And it seems that the plug-in doesn't work with the blinks.csv file. And I've found the plugin Post hoc blink detection, but I don't know if it allow me to export the images of the world video during the blink https://gist.github.com/papr/40ba7d99f572bc0fe388a81aa2f87424.
Hi! Pupil Player does not support blink detection for Pupil Invisible recordings. It is based on pupil confidence which cannot be calculated for this type of recordings. You can download blink data from Pupil Cloud though!
Never the less, even the Post hoc blink dectection allow me to do this. I've problem to make it works correctly. I've ERROR message : player - [WARNING] plugin: Failed to load 'batch_exporter'. Reason: 'No module named 'exporter'' player - [WARNING] plugin: Failed to load 'extract_blinks'. Reason: 'No module named 'pandas'' But I've already installed exporter and pandas on my computer...
Does anyone have an idea to solve those problems ? I'm not devlopper and I don't know python, so that's part of the problem. But if any of you have a teacher soul and the patience to help me it could be nice.
Hey @papr , I believe that pye3d uses optimization (ceres solver) to fit the 3D eye model. Can you point me to where this happens in the pye3d package?
That was only the case for the old 3d detector. Pye3d performs a closed-form optimization.
...and is the optimization of reprojection error?
For some context, we've now played enough with the entire pipeline to develop the strong opinion that this is the step that's contributing the majority of error. Not the binocular model fit.
By binocular model fit you are referring to the gaze calibration?
Not pupil detection.
By that you are referring to the 2d detection?
(although that may be a related issue, our models aren't improving things)
That's the closed Form optimization
Thank you, Pablo!
One more idea to drop ...
We've found that the precision of the final gaze estimate is relatively poor when the model is not frozen. The high-frequency sample-to-sample updating seems to be the issue. We could play with the paramters and calm it, perhaps, but my first inclination is that the updates should not be applied directly to the 3D eye pose center.
The short term model only affects the gaze direction, not the sphere center
Instead, I would consider applying them to an attractor towards which the eyeball center accelerates.
perhaps in a PD-controller-like fashion. This is just one of many possible ways to low-pass filter those updates.
Hello, is it possible to plot x,y coordinate on the screen from norm_pos_x and norm_pos_y?
Ho, not directly, since norm_pos_x/y are in scene camera coordinates by default, not in screen coordinates. What you need is https://docs.pupil-labs.com/core/software/pupil-capture/#surface-tracking
hey @papr thanks for your quick response on github. I just installed C-Make version 3.25 (via snap) and still unable to build pyuvc 😦
I think the current error is:|
`CMake Error at CMakeLists.txt:103 (target_link_libraries): Target "uvc_bindings" links to:
PLLibUVC::uvc
but the target was not found. Possible reasons include:
`
Let's move this discussion into #1039477832440631366
I have POV view as well from the main camera of pupil capture. Can I use that overlay dots on the screenshot showing where the subject was looking at?
Where is the difference between POV and main camera for you?
The issue is that pov/main camera are not static in relation to the screenshot. But gaze is estimated in the former, not the latter. You need an additional mapping step. And this is what surface tracking does for you.
By POV I mean this!
I see. So technically speaking there is no workaround for what I want to do?
There is, but you need to use the surface tracking.