A New Approach for Building a Saliency System for Eye Fixation Prediction

The purpose of this project is to build a saliency system that is generalizable and can predict human eye fixation.

• Generative Adversarial Networks
• Autoencoders
• Multiple-scale processing

• Python
• TensorFlow

Here is a link to a presentation to the project: https://informatikunihamburgde-my.sharepoint.com/:b:/g/personal/shahd_safarani_informatik_uni-hamburg_de/EUxaFbK-AuFDoiQjUmfORHEBga4sgqxlRstEaQWw8TNdqQ?e=hbyGXk

SALICON dataset is used for training. Link to download the dataset: https://informatikunihamburgde-my.sharepoint.com/:f:/g/personal/shahd_safarani_informatik_uni-hamburg_de/EvdZ1uO6wJlAmFuyl7sdZo0BU3HP06lQAo_EdNewENGQyg?e=aSqC2j

Link to trained models : https://informatikunihamburgde-my.sharepoint.com/:f:/g/personal/shahd_safarani_informatik_uni-hamburg_de/Ekr_EDZdTsdHt86O0wNV1PQBfaRmwI3Ds1amMpie539m6A?e=qJVfkU

Link to TensorBoard summaries: https://informatikunihamburgde-my.sharepoint.com/:f:/g/personal/shahd_safarani_informatik_uni-hamburg_de/EiwS2jSNwqRCgzJPapdHN4MBnxaKVw7Z6dBqy8_HvNCGqQ?e=wVboXZ

Results on test images of MIT300 benchmark already submitted and waiting for evaluation.

Please in case of any quetions, contact info: [email protected]