Official repository for the IEEE Access paper: IEEEXplore | ArXiv
Human perceptual priors have shown promise in saliency-guided deep learning training, particularly in the domain of iris presentation attack detection (PAD). Common saliency approaches include hand annotations obtained via mouse clicks and eye gaze heatmaps derived from eye tracking data. However, the most effective form of human saliency for raising generalization to unknown attack classes in iris PAD remains under-explored. In this paper, we conduct a series of experiments comparing hand annotations, eye tracking heatmaps, segmentation masks, and foundation model embeddings to a state-of-the-art deep learning-based baseline on the task of unknown attack type classification for iris PAD. Results in a leave-one-attack-type out paradigm indicate that denoised eye tracking heatmaps show the best generalization improvement over cross entropy in Attack Presentation Classification Error Rate (APCER) at Bona Fide Presentation Classification Error Rate (BPCER) of 1%. Along with this paper, we offer trained models, code, and saliency maps for reproducibility and to facilitate follow-up research efforts.
At a high level, the dataset is a JSON file where each object includes:
- A reference to the original iris image
- Ground truth label
- Attack type category
- Links to saliency maps for each tested configuration for the image
The dataset is organized as a list of JSON objects where each object refers to an iris sample from the dataset described in the paper. Each JSON object contains a reference to the iris sample and the attack type the sample represents with Live indicating Bonafide, or Spoof indicating some type of presentation attack category. Additionally there is a dictionary of image links that correspond to the different saliency map configurations that are associated with this image, i.e. "Denoised_Initial_ET": "Denoised_Initial_ET/9_5_blended.png" is a reference to the de-noised initial eye tracking saliency map for the image 9_5.png that was tested during the experiments.
{
"irisImageLink": "9_5.png",
"label": "Live",
"attackType": "Live",
"saliencyMaps": {
"Denoised_Initial_ET": "Denoised_Initial_ET/9_5_blended.png",
"Denoised_Full_ET": "Denoised_Full_ET/9_5_blended.png",
"Initial_Eye_Tracking": "Initial_Eye_Tracking/9_5_blended.png",
"Full_Eye_Tracking": "Full_Eye_Tracking/9_5_blended.png",
"Segmentation_Masks": "Segmentation_Masks/9_5.png",
"Hand_Annotations_Low_Entropy": "Hand_Annotations_Low_Entropy/9_5_blended.png",
"Hand_Annotations_Equal_Entropy": "Hand_Annotations_Equal_Entropy/9_5_blended.png",
"Hand_Annotations_High_Entropy": "Hand_Annotations_High_Entropy/9_5_blended.png"
}
}Instructions on how to obtain a copy of the dataset can be found at the Notre Dame's Computer Vision Research Lab webpage (VISER Dataset). Any questions can be directed to Adam Czajka at aczajka@nd.edu.
@article{dowling2026viser,
title={VISER: Visually-Informed System for Enhanced Robustness in Open-Set Iris Presentation Attack Detection},
author={Dowling, Byron and Piland, Jacob and Frederick, Eleanor and Sweet, Christopher and Czajka, Adam},
journal={arXiv preprint arXiv:2603.17859},
year={2026}
}
This work was supported by the U.S. Department of Defense (Contract No. W52P1J-20-9-3009) and by the National Science Foundation (Grant No. 2237880). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation, the U.S. Department of Defense or the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for Government purposes, notwithstanding any copyright notation here on.