Eye Tracking Comparison

This project procides:

• Visual Test Battery with 9 different Eye Tracking Tasks (Matlab/PsychophysicsToolbox-3)
• Preprocessing Pipeline for Pupil Labs glasses & EyeLink 1000. Modular for other eye trackers (Python)
• Analysis Scripts for Ehinger, Groß et al 2019 (Python)
• Makefile for all requirements, including Pupil Labs Software WITHOUT root/sudo necessity (see below for restrictions)
• Pupil Labs Eye Videos + World Video of 15 subjects + concurrent EyeLink data at figshare

Instructions

Get the project GIT and initialize the submodules

git clone https://github.com/behinger/etcomp
git submodule update --init

This Makefile generates a virtualenv, downloads all required python modules, installs all requisits for Pupil Labs. This includes opencv, it my take quite a while!

make

Visual Test Battery

Acuity TEST

We used this acuity test. Be sure to setup monitor size correctly!

Run Experiment

The experiment can be run by ./experiment/ETcomp.m

Ubuntu: The experiment communicates with Pupil Labs using ZMQ. To get this to work, you have to load a library before running matlab:

LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libstdc++.so.6 matlab

Python Preprocessing Pipeline

Update 2021-12-10: You could also check out this fork for a (potentially) cleaned up version. Haven't checked or tried it personally - citation would still be appreciated :-)

You can run a subject using the preprocessing pipeline (from the folder "code"). We took some care to make it modular, but the pipeline should be more thought as a starting base and needs modification for your own setup.

import functions.et_preprocess
from functions.detect_events import make_blinks,make_saccades,make_fixations
data = preprocess_et(et='pl', datapath='/net/store/nbp/projects/etcomp/sub-1/raw', eventfunctions=(make_blinks,make_saccades,make_fixations))

• The datapath should be a pupil labs folder (if et='pl'), or a folder with an eyelink-EDF file (if et='el')
• The eventfunctions are modular, here we use the default ones implemented by us, but you could simply replace them with other functions. The eventfunctions are executed sequentially.
• We are using loggers for messages, try to use the "debug" flag if you want more information!

Python Analysis Pipeline

Our analyses (and all results included in the paper) can be found in the notebook: code/main.ipynb

Further analyses can be found in the capital notebooks eg. SMOOTH.ipynb, GRID.ipynb. Functions for each tasks are in SMOOTH.py, or GRID.py

Makefile

We made great efforts to not need root rights but still be able to use pupil labs from source. The GUI is non-functional. pyAV somehow is a newer version and we need to fix some things in pupillabs (forked version is linked)

You need these packages definitely installed using sudo/ root (there might be more):

automake
cmake
pkgconfig
python3-dev
libglew-dev <-- this one is usually not installed
xorg-dev libglu1-mesa-dev <-- needed for libglew

(In principle you can also try to compile glew, but I could not manage properly.)

Add Paths

After installation run make export-paths

and everytime before running Spyder or Python, you need to run these to add some dependencies for Python to use.

You can put these in your ~./bashrc so that everytime you run a terminal, you will also have the paths automatically (highly recommend)

EDFread

Pyedfread also needs a dependencie from SR-research (libedfapi.so), which is not publicly available. Checkout the Sr research forum!

Old instructions

These instructions here are notes on how to instlal different versions of Pupil Labs glasses.

How to start the compiled binary of pupil player / capture locally under ubuntu:

• Get the linux release zip file from https://github.com/pupil-labs/pupil/releases
• unzip the .deb files from it
• run dpkg -x pupil_capture_linux_os_x64_v1.6.13.deb /folder/to/be/extracted/to
• Optional: Copy the file to a more useful directory e.g. /net/store/nbp/users/yourRZname
• go to the folder cd /folder/to/be/extracted/to and run ./pupil_player

What follows now is old installation instruction, they were merged to make install

Python EDFREAD

• go to lib/pyedfread
• make new folder: lib
• copy /net/store/nbp/projects/lib/edfread/build/linux64/libedfapi.so to lib/libedfapi.so
• make new folder include
• copy /net/store/nbp/projects/lib/edfread/*.h to lib/include/*.h
• activate your virtualenv
• run python setup.py install

install opencv

• compile opencv3 as pupil-labs

  git clone https://github.com/itseez/opencv
  cd opencv
  mkdir build &&  cd build
  cmake -D CMAKE_BUILD_TYPE=RELEASE -D BUILD_TBB=ON -D WITH_TBB=ON -D WITH_CUDA=OFF -D BUILD_opencv_python2=OFF -DBUILD_opencv_python3=ON  -D CMAKE_INSTALL_PREFIX='XXX/opencv-build'
  make -j4
  make install
  

isntalling pyav under linux

This one was difficult

• get ffmpeg3 (only v.2 is installed by defaul)

• I used https://launchpad.net/ubuntu/+archive/primary/+files/ffmpeg_3.3.4.orig.tar.xz

• Unzip it

• you likely need yasm, but its straight forward to build

  • git clone https://github.com/yasm/yasm
  • cd yasm
  • sh autoconfig.sh
  • ./configure --prefix=../yasm-build # or cmake?
  • make yasm
  • make install
  • export PATH=$PATH:/net/store/nbp/users/behinger/tmp/pupil_src_test/yasm-build/bin
  • should be done, I did not have trouble here at all

• run ./configure --prefix=../ffmpeg-build --enable-shared --enable-pic --cc="gcc -m64 -fPIC" --extra-cflags="-fPIC" (the enabled shared is key, else we later get errors compiling av)

• run make and then make install

• add the newly build ffmpeg libraries to the pkg-config path: export PKG_CONFIG_PATH=:folder/to/install/ffmpeg/:$PKG_CONFIG_PATH

• in your python3 venv, run "pip install av'

• you always have to add export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/net/store/nbp/users/behinger/tmp/pupil_src_test/ffmpeg-build/lib/ before import av (you should else get an error ImportError: libavfilter.so.6: cannot open shared object file: No such file or directory)

• I often ran into the problem avformat_open_input while installing pip av. Not sure what solved this. I cloned the pyav-git, and removed both instances of avformat_open_input (just commented them out). Compiled then successfully, put them back in, and compiled again - now it worked... strange

• start python and type import av - if it works, I'm happy for you... took me 2.5h