SPGen: Stochastic scanpath generation for paintings using unsupervised domain adaptation

2026-02-25

AI Summary

This work proposes SPGen, a novel model for accurately predicting human scanpaths when viewing paintings. Built upon a fully convolutional network, SPGen integrates a differentiable fixation selection mechanism and a learnable Gaussian prior to capture natural viewing preferences. It is the first approach to jointly incorporate unsupervised domain adaptation and a stochastic noise sampler into scanpath generation for artworks, effectively bridging the domain gap between natural images and artistic paintings without requiring labeled data in the target domain. The method significantly enhances both the realism and diversity of generated scanpaths, outperforming existing approaches across multiple evaluation metrics. By doing so, SPGen provides an effective computational tool for advancing research in art perception and supporting digital heritage applications.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Natural Language Processing: Generation
• Cognitive Modeling & Cognitive Systems: Computational Creativity

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Web Mining and Content Analysis: Web data generation and simulation
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Understanding human visual attention is key to preserving cultural heritage We introduce SPGen a novel deep learning model to predict scanpaths the sequence of eye movementswhen viewers observe paintings. Our architecture uses a Fully Convolutional Neural Network FCNN with differentiable fixation selection and learnable Gaussian priors to simulate natural viewing biases To address the domain gap between photographs and artworks we employ unsupervised domain adaptation via a gradient reversal layer allowing the model to transfer knowledge from natural scenes to paintings Furthermore a random noise sampler models the inherent stochasticity of eyetracking data. Extensive testing shows SPGen outperforms existing methods offering a powerful tool to analyze gaze behavior and advance the preservation and appreciation of artistic treasures.

Problem

• scanpath prediction
• visual attention
• paintings
• eye movements
• cultural heritage

Innovation

• stochastic scanpath generation
• unsupervised domain adaptation
• fully convolutional neural network
• gradient reversal layer
• visual attention modeling