Skip to content

Eye gaze prediction using recurrent convolutional neural networks. This repo contains the code I wrote (Tensorflow+Keras) to train my models for Facebook's OpenEDS2020 eye gaze prediction challenge.

Notifications You must be signed in to change notification settings

isayasMatter/Gaze-Prediction

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

17 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Eye gaze prediction - OpenEDS 2020 Challenge

This repo contains the code I wrote for the OpenEDS 2020 Challenge. You can learn more about the challenge here. The challenge had two tracks, one of which was the gaze prediction challenge. The challange's main goals were:

  • to predict future gaze locations based on previously estimated eye gaze vectors
  • to predict future gaze locations based on previously estimated gaze vectors while additionally leveraging spatio-temporal information encoded in sequence of previously recorded eye-images.

My submission

I used Tensorflow with Keras to implement a recurrent convolutional neural network. The network consists of two parts: 1) a Resnet (various variants) based part that learn features from each eye image, and 2) a many-to-one recurrent neural network based temporal part that predicts gaze positions.

My submission scored way better than the baseline score and very close to the winning submission. Feel free to use the code as you see fit and shoot me an email if you have any questions about the code.

About

Eye gaze prediction using recurrent convolutional neural networks. This repo contains the code I wrote (Tensorflow+Keras) to train my models for Facebook's OpenEDS2020 eye gaze prediction challenge.

Topics

Resources

Stars

Watchers

Forks

Languages