Laser-ScanPath: GPT-4o Annotated Visual Scanpaths for 269,773 Images

269,773 samples of GPT-4o generated scan-path annotations for training the LASER model. The dataset, created by author 'wybb', references images from six Visual-CoT subsets but does not include the images themselves. It was last updated on the platform in April 2026.

Use Cases

• Training visual attention models based on synthetic eye-tracking data.
• Benchmarking scan-path prediction algorithms using the GPT-4o generated annotations.
• Studying the relationship between image content and AI-predicted visual focus across six distinct visual question-answering datasets.

Strengths

• 269,773 annotated samples provide a substantial base for model training.
• Annotations are generated by GPT-4o, a state-of-the-art multimodal model.
• Links image data from six established visual reasoning datasets: flickr30k, gqa, openimages, v7w, cub, and vsr.

Limitations

• The core image files are not included; they must be downloaded and organized separately from Visual-CoT.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known, but other metadata like license, file formats, and sample data are unavailable.

Provenance

Source

huggingface, author wybb

Collection Method

Annotations generated by GPT-4o for LASER training.

Time Range

null

Freshness

Last updated 2026-04-08 14:47:12.

Geography

null