Where do people look when they view a painting, and how does that change across artistic styles? AVAtt, introduced at ETRA 2024, is a dataset of human visual attention over paintings spanning diverse styles, built to support attention and scanpath research in the art domain.

Most saliency and attention datasets are grounded in natural photographs, which leaves art conspicuously underrepresented. Paintings carry deliberate compositional choices, abstraction, and stylistic variation that natural-image models were never designed to capture, so studying attention on art requires data that reflects that diversity.

AVAtt provides eye-tracking-based attention measurements across a range of painting styles, giving researchers a resource to analyze how gaze behaves when style, composition, and artistic intent come into play. The emphasis on stylistic diversity is central to its value.

We hope AVAtt encourages more work at the intersection of computational attention and the arts: from benchmarking saliency models on non-photographic content to studying scanpaths shaped by artistic form. This was joint work with M. Tliba, A. Chetouani, and A. Bruno.

The dataset and accompanying code are released openly as AVAtt for the community to build on.

See the paper for the full methodology and results.