DATASHEET.md

PolyCAT Datasheet

Following the "Datasheets for Datasets" framework (Gebru et al., 2021).

Motivation

For what purpose was the dataset created? PolyCAT was created to study how geometric viewing constraints (polygon-shaped apertures) affect visual attention allocation on natural images. It enables saliency prediction research under non-rectangular viewing conditions.

Who created the dataset and on behalf of which entity? Created by researchers at [institution name]. Funded by [funding source].

Who funded the creation of the dataset? [To be filled]

Composition

What do the instances that comprise the dataset represent? Each instance is one eye-tracking trial: a participant's gaze behavior while viewing a CAT2000 image through a polygon aperture for 4 seconds.

How many instances are there in total? ~21,000 trials across 30 participants (702 trials per fully included participant, 2 parts x 351 trials). All 30 participants are fully included with both parts.

Does the dataset contain all possible instances or is it a sample? It is a complete recording of all trials for all included participants. 600 of the ~2000 CAT2000 images were used.

What data does each instance consist of?

• Sample-level gaze coordinates at 500 Hz (binocular)
• Fixation events with position, duration, and timing
• Trial metadata (stimulus, polygon, cue position)
• Session metadata (calibration, validation)

Is there a label or target associated with each instance? No explicit labels. The gaze data itself can be used to derive saliency maps.

Is any information missing from individual instances? Some trials have tracking loss (blinks, look-away). Quality metrics are provided per-trial.

Are there any errors, sources of noise, or redundancies?

• EyeLink tracking noise (~0.25-0.5 deg typical accuracy)
• Blinks cause data gaps
• Some edge fixations may be at aperture boundaries

Collection Process

How was the data associated with each instance acquired? Eye movements were recorded using an EyeLink 1000+ eye tracker while participants viewed stimuli on a 27" 4K monitor.

What mechanisms or procedures were used to collect the data? Custom PsychoPy experiment with EyeLink integration. 13-point calibration before each block, drift check before each trial.

Who was involved in the data collection process? Trained experimenters supervised each session. Participants were university students.

Over what timeframe was the data collected? January 15 - February 25, 2026.

Were any ethical review processes conducted? Yes. The study was approved by the institutional ethics review board. All participants provided written informed consent.

Preprocessing

Was any preprocessing/cleaning/labeling of the data done?

• Raw EDF files were converted to CSV/TSV format
• Fixations were extracted using EyeLink's built-in detection algorithm
• Split sessions were concatenated using trial UIDs
• Quality metrics were computed per trial and participant

Was the "raw" data saved in addition to the preprocessed/cleaned/labeled data? Yes. The original session logs and EDF files are preserved in data/raw_output_data/.

Uses

Has the dataset been used for any tasks already? Initial analyses are presented in the accompanying ETRA 2026 paper.

What (other) tasks could the dataset be used for?

• Saliency prediction with aperture constraints
• Scanpath modeling and prediction
• Center bias analysis under geometric constraints
• Polygon geometry effects on visual exploration

Is there anything about the composition of the dataset or the way it was collected that might impact future uses? The polygon apertures create a unique viewing constraint not present in standard saliency datasets. Models should account for the masked regions.

Distribution

How will the dataset be distributed? Via the university lab's SharePoint repository. A permanent link will be provided in the published paper.

When was the dataset first released? 2026 (ETRA 2026 conference).

What license is the dataset distributed under? CC BY 4.0. Note that the CAT2000 images are subject to their original license.

Maintenance

Who is supporting/hosting/maintaining the dataset? [Institution name and contact information]

Will the dataset be updated? Version updates will be documented with changelogs.

How can the owner/curator/manager of the dataset be contacted? [Contact email]