Identifying from Gaze Fixations the Desired Objects and Faces

Current version of main.py uses tensorflow object detection network trained on icub faces

So you need to get the icub-face-detection repo to make it work

Table of Contents

• Setup
• Instructions
• Installation
• Dependencies
• Possible Issues

Setup

Running in Online mode

The software processes works in real-time by processing the information of the PupilLabs head-mpinted eye-tracker online and writting to a YARP port for communication with a humanoid robot (iCub). The whole process is interconnected by a LSL network.

Fig.1 - Diagram illustrating the connections between the different modules that make up the communication of the human eye gaze to the robot fixations. The first module is related to the software that acquires the data from the eye tracker - Capture by Pupil Labs. From this module we collect the 2D fixation point of the subject's gaze projected onto the world view camera on the eye tracker. The stream of the world view camera, together with 2D gaze fixations through LSL network, is sent to idGaze software module to track the relevant fixations.

Running in Offline mode

The software processes recorded data exported from the PupilLabs head-mounted eye-tracker.

Fig.2 - Diagram illustrating the conversion from raw Pupil Labs data to labelled gaze fixations

Instructions

• There are two conda virtual environments which were tested and was running (pupilos and pupilos-10). Pupilos was working for tensorflow 1.9 and CUDA-8.0 and pupilos-10 was working for tensorflow 2.7 and CUDA-11.2

1. add anaconda

source activate pupilos***

2. install tensorflow models - object detection
3. add package utils (follow dependencies)
4. run main.py

python3 main.py

or main_offline.py

python3 main_offline.py

Dependencies

Tensorflow Object Detection Model - You need to use the following packages from utils:

from utils import label_map_util
from utils import visualization_utils as vis_util

to import utils you need to install tensorflow with gpu then get the models of tensorflow for object recognition to recognize the import you need the following (after you have followed the instructions on how to install tensorflow models)

cd software/tensorflow/models/research
export PYTHONPATH=$PYTHONPATH:$(pwd)/slim
echo $PYTHONPATH 
export PYTHONPATH=$PYTHONPATH:$(pwd):$(pwd)/object_detection 

Issues

If you find this problem when using codeblocks:

pthread_create@2.25 

The solution is to add the thread library to both the compiler and linker. You can find an example in this link.

If you find that it does not recognize libYARP_init.so.3 even though it recognizes libYARP_init.so. A good quick workaround is to add the path to the yarp libs to the Search directories. You can find an example in this link. You have to go to project->Build options...->Search directories->Linker and add the directory there "../../../../middleware/yarp/build/lib"

If you cannot find shared library (e.g. liblsl64.so.1.2.0) it works if you right click project->Build options...->Search directories->Linker and add the directory there. The directory where liblsl64.so.1.2.0 is for example