fdiss 0.1.0

Foveal Disc IoU Scanpath Score - A biologically grounded scanpath similarity metric

FDISS

Foveal Disc IoU Scanpath Score

A biologically grounded scanpath similarity metric that models fixations as foveal discs and computes similarity via bidirectional nearest-neighbour IoU matching.

Installation

pip install fdiss

Quick Start

import numpy as np
from fdiss import FDISS

# Initialize the metric
fdiss = FDISS()

# Two scanpaths (x, y coordinates)
S1 = np.array([[100, 200], [150, 250], [300, 150]])
S2 = np.array([[105, 205], [155, 255], [320, 160]])

# Compute foveal radius (45px = 1 degree at 45 px/degree)
r = FDISS.compute_foveal_radius(px_per_degree=45.0, degrees=1.0)

# Evaluate similarity
result = fdiss.evaluate(S1, S2, r)

print(f"FDISS: {result['fdiss']:.4f}")
print(f"Precision: {result['precision']:.4f}")
print(f"Recall: {result['recall']:.4f}")

Features

• Biologically Grounded: Models fixations as foveal discs (~1-2° visual angle)
• Interpretable: Returns Precision, Recall, and F1 (FDISS) scores
• Fast: O(nm) complexity with vectorized NumPy operations
• No Parameters at Runtime: Foveal radius computed once from viewing geometry

Method

FDISS represents each fixation as a foveal disc with radius r, then computes similarity using:

1. Disc IoU: Closed-form intersection-over-union of two equal-radius discs
2. Bidirectional Matching: Each fixation matched to its best-overlapping counterpart
3. Precision/Recall: Coverage from both directions
4. FDISS: Harmonic mean of Precision and Recall

Citation

If you use FDISS in your research, please cite:

@article{kerkouri2026fdiss,
  title={FDISS: Foveal Disc IoU Scanpath Score},
  author={Kerkouri, Mohammed Amine and Tliba, Marouane and Chetouani, Aladine},
  journal={ETRA '26},
  year={2026}
}

License

MIT License