CAT2000: A Large Scale Fixation Dataset for Boosting Saliency Research

Paper

CAT2000: A Large Scale Fixation Dataset for Boosting Saliency Research

Published May 14, 2015 · A. Borji, L. Itti

ArXiv

UNKNOWN SJR score

272

Citations

31

Influential Citations

Semantic Scholar

Abstract

Saliency modeling has been an active research area in computer vision for about two decades. Existing state of the art models perform very well in predicting where people look in natural scenes. There is, however, the risk that these models may have been overfitting themselves to available small scale biased datasets, thus trapping the progress in a local minimum. To gain a deeper insight regarding current issues in saliency modeling and to better gauge progress, we recorded eye movements of 120 observers while they freely viewed a large number of naturalistic and artificial images. Our stimuli includes 4000 images; 200 from each of 20 categories covering different types of scenes such as Cartoons, Art, Objects, Low resolution images, Indoor, Outdoor, Jumbled, Random, and Line drawings. We analyze some basic properties of this dataset and compare some successful models. We believe that our dataset opens new challenges for the next generation of saliency models and helps conduct behavioral studies on bottom-up visual attention.

Highly Cited

Preprint

Study Snapshot

Key takeawayOur large-scale fixation dataset CAT2000 offers new challenges for next-generation saliency models and helps conduct behavioral studies on bottom-up visual attention.

PopulationOlder adults (50-71 years)

Sample size24

MethodsObservational

OutcomesBody Mass Index projections

ResultsSocial networks mitigate obesity in older groups.

Sign up to use Study Snapshot

Consensus is limited without an account. Create an account or sign in to get more searches and use the Study Snapshot.

Create an account

Paper

CAT2000: A Large Scale Fixation Dataset for Boosting Saliency Research

Sign up to use Study Snapshot

References

How do humans sketch objects?

Human beings can correctly identify object categories in sketches 73% of the time, and our computational model can identify unknown sketches with 56% accuracy.

On the relationship between optical variability, visual saliency, and eye fixations: a computational approach.

This study proposes a hierarchical definition of optical variability that links physical magnitudes to visual saliency, and suggests that contextual adaptation mechanisms aim to ensure invariance of behavior elicited by optical variability in the visual system.

Human action recognition by learning bases of action attributes and parts

Our method effectively recognizes human actions in still images by learning sparse bases of action attributes and parts, achieving state-of-the-art classification performance on the PASCAL and Stanford 40 Actions datasets.

Unbiased look at dataset bias

This paper compares popular object recognition datasets to evaluate their biases, generalization, closed-world assumptions, and sample value, suggesting improvements in dataset collection and algorithm evaluation protocols.

Fixations on low-resolution images.

Human fixations are more consistent at higher resolutions, with a stronger bias towards the center, and consistency remains relatively constant despite the spread of fixations away from the center.

Citations

RClicks: Realistic Click Simulation for Benchmarking Interactive Segmentation

RClicks benchmark accurately simulates real-user clicks, revealing that interactive segmentation methods may perform worse in real-world usage and are often not robust.

VISTA: A Visual and Textual Attention Dataset for Interpreting Multimodal Models

This study presents an image-text aligned human visual attention dataset, enabling better understanding of Vision and Language Models' decision-making processes and improving model transparency and interpretability.

Single cone photoreceptors experience global image statistics through gaze shifts

Local adaptation in cone photoreceptors can effectively compensate for illumination changes, maintaining stable local perception of scene color despite illumination changes.

Data Augmentation via Latent Diffusion for Saliency Prediction

Our novel data augmentation method improves saliency prediction models by incorporating photometric and semantic attributes, preserving scene complexity, and aligning with human visual attention patterns.

SUM: Saliency Unification through Mamba for Visual Attention Modeling

SUM is a unified model for visual attention modeling that adapts to diverse image types and consistently outperforms existing models in various applications.

GESCAM : A Dataset and Method on Gaze Estimation for Classroom Attention Measurement

GESCAM, a synthetic dataset, effectively measures classroom attention using gaze target detection, providing insights into human attention in diverse contexts.

Varying Manifolds in Diffusion: From Time-varying Geometries to Visual Saliency

Our generation curve matching algorithm improves image manipulation results by controlling the local deformation of manifolds in deep generative models.