Project Overview
This repository implements three saliency prediction models (CovSal, GBVS, and FES) to generate saliency maps and evaluate their alignment with human fixation patterns in Typically Developed (TD) and Autism Spectrum Disorder (ASD) groups.

Paper & Source: CovSal Project

Saliency Map Prediction Code: Download Here

1. Download the code file from the provided link.
2. Extract the ZIP file; it will create a folder named saliency.
3. Navigate to the saliency folder (ignore _MACOSX if present).
4. Open MATLAB and create a .m file.
5. Use saliencymap.m to generate saliency maps.

1. Open the CovSal folder inside the repository.
2. Navigate to the saliency folder.
3. Run generate_saliency_maps.m in MATLAB to generate saliency maps for selected folders.
4. Modify the folder directory in the script to match your dataset location.

• Run perform_evaluation.m in MATLAB to evaluate the generated saliency maps.

Paper & Source: GBVS Project
Paper Title: Image Saliency Estimation via Random Walk Guided by Informativeness and Latent Signal Correlations

Saliency Map Prediction Code: Download Here

1. Download the code file from the provided link.
2. Extract the ZIP file; it will create a folder named corSal.
3. Open MATLAB and create a .m file.
4. Navigate to external/gbvs/gbvs.m to generate saliency maps.

1. Open the CovSal folder inside the repository.
2. Navigate to the corSal folder.
3. Run generate_saliency_map.m in MATLAB to generate saliency maps for selected folders.
4. Modify the folder directory in the script to match your dataset location.

• Open the corSal folder and run perform_evaluation.m in MATLAB.

Paper & Source: FES Paper

Saliency Map Prediction Code: GitHub Repository

1. Download the code from the provided GitHub link.
2. Extract the ZIP file; it will create a folder named FES-master.
3. Open MATLAB and create a .m file.
4. Run calculateImageSaliency.m to generate saliency maps.

1. Open the FES-master folder inside the repository.
2. Run generate_saliency_map.m in MATLAB to generate saliency maps for selected folders.
3. Modify the folder directory in the script to match your dataset location.

• Open the FES-master folder and run perform_evaluation.m in MATLAB.

The saliency maps generated by each model are evaluated using the following metrics:

• AUC_Borji (Area Under Curve - Borji)
• AUC_Judd (Area Under Curve - Judd)
• AUC_shuffled (Shuffled AUC)
• CC (Correlation Coefficient)
• EMD (Earth Mover’s Distance)
• Info Gain (Information Gain)
• KLdiv (Kullback-Leibler Divergence)
• NSS (Normalized Scanpath Saliency)

• Use perform_evaluation.m in each folder (saliency, corSal, FES-master).
• The same evaluation script is used for all models.
• Ensure the paths are set correctly:
  • td_fixation_folder and asd_fixation_folder should point to the respective fixation files.
  • prediction_folder should be set to the saliency map folder generated by each model.

The dataset used in this project can be downloaded from: Dataset Link

Duan H, Zhai G, Min X, Che Z, Fang Y, Yang X, Gutiérrez J, Callet PL. A dataset of eye movements for the children with autism spectrum disorder. In Proceedings of the 10th ACM Multimedia Systems Conference 2019 Jun 18 (pp. 255-260).

• CovSal, GBVS, and FES implementations are sourced from the respective research papers.
• We acknowledge the original authors for their contributions to saliency map research.