Home

Awesome

Unveiling the Truth (IEEE Signal Processing Letters 2024)

Exploring Human Gaze Patterns in Fake Images

Giuseppe Cartella, Vittorio Cuculo, Marcella Cornia, Rita Cucchiara

This is the official repository for the paper "Unveiling the Truth: Exploring Human Gaze Patterns in Fake Images".

📣 Latest News 📣

Overview

<p align="center"> <img src="teaser.jpg" style="max-width:500px"> </p>

Download the dataset

Dataset download is available here.

The dataset should be structured as follows:

|-dataset/
    |-images/
        |-original/
            |- <img_id_1>.jpg
            |- ...
        |-pix2pix_magicbrush/
            |- <img_id_1>.jpg
            |- ...
        |-semantic_agnostic/
            |- <img_id_1>.jpg
            |- ...
        |-semantic_aware/
            |- <img_id_1>.jpg
            |- ...
    |-masks/
        |-semantic_agnostic/
            |- <img_id_1>.png
            |- ..
        |-semantic_aware/
            |- <img_id_1>.png
            |- ...

    |-gaze_data.pickle
    |-ip2p_edit_instruction_ADE_subset.json
    |-ip2p_edit_instruction_COCO_subset.json
    |-ip2p_edit_instruction_LHQ_subset.json
    |-semantic_aware_prompt_ADE_subset.json
    |-semantic_aware_prompt_COCO_subset.json
    |-semantic_aware_prompt_LHQ_subset.json
    |-readme.txt

Please refer to the readme.txt file for all details.

How to Visualize Scanpaths

As a first step, move the dataset into the project directory.

To visualize a scanpath run the following command:

python visualize_scanpath.py --img_id COCO_000000229478 --user=4 --type=semantic_aware

By default, two output files are generated:

Red and blue circles represent the first and latest fixation of the plotted scanpath, respectively.

TODO