README.md

---
license: cc-by-4.0
size_categories:
- n<1K
pretty_name: OpenSAL360
tags:
- video
- omnidirectional
- saliency
- crowdsourcing
task_categories:
- other
---

# OpenSAL360: Open-Source Crowdsourcing Platform for Omnidirectional Video Saliency Collection
[![OpenSAL360 Code](https://img.shields.io/badge/OpenSAL360-Code-brightgreen?style=for-the-badge)](https://github.com/msu-video-group/OpenSAL360)
![image](https://cdn-uploads.huggingface.co/production/uploads/623a3b20fa4890c51b04cba7/qyOoHGqxr06O16i9LyI8w.png)

**Paper:** [OpenSAL360: Open-Source Crowdsourcing Platform for Omnidirectional Video Saliency Collection](https://huggingface.co/papers/2609.21480)

## Dataset
**OpenSAL360** sample dataset is the largest omnidirectional video saliency dataset with the following key features:
* Large scale: **500** videos with mean **18.1s** duration;
* Fixations from **>2000** observers (**>84** per video);
* **Audio** track saved and played to observers;
* High resolution: all streams are **3840x1920**;
* Diverse content from **YouTube**;
* License: **CC-BY**;

### File Structure

1) `videos` — 500 mp4 omnidirectional videos in ERP, 30 FPS videos with audio streams;

2) `saliency` — 500 mp4 almost losslessly (crf 0, 10bit, min-max normalized) compressed continuous saliency maps videos;

3) `fixations` — 500 json files with per-frame fixation coordinates, from which saliency maps were obtained;

4) `metadata.json` — meta information about each video (e.g. license, source URL, etc.);