OpenSAL360 is currently the largest comprehensive video saliency dataset, designed to support research in visual attention, saliency prediction, and multimodal video analysis. It is widely used in various research and engineering fields such as panoramic video understanding, visual attention modeling, saliency prediction algorithm evaluation, multimodal perception research, and VR/AR interactive system design. This dataset contains 500 diverse panoramic videos from YouTube, with an average duration of 18.1 seconds. All video streams have a resolution of 3,840 x 1,920. The data was annotated by more than 2,000 observers, and each video contains an average of more than 84 eye-tracking fixations, while the original audio tracks are fully preserved.

Data Structure

• Videos: Includes 500 MP4 videos, full-range, 30 FPS, with audio streaming.
• Saliency: 500 near-lossless compressed saliency maps (video)
• Fixations: 500 JSON files, each containing gaze coordinates, which can be used to generate saliency maps.
• metadata.json: Records metadata information for each video, including license, source URL, etc.