---
license: apache-2.0
language:
- en
tags:
- fMRI
- EEG
- neuroscience
- brain-imaging
- science
- huggingscience
size_categories:
- 100K<n<1M
task_categories:
- text-to-video
- image-to-video
pretty_name: CineBrain
---

# [CVPR 2026] CineBrain: A Large-Scale Multi-Modal Brain Dataset During Naturalistic Audiovisual Narrative Processing
[![ArXiv](https://img.shields.io/badge/ArXiv-2503.06940-b31b1b.svg?logo=arXiv)](https://arxiv.org/abs/2503.06940)
[![Code](https://img.shields.io/badge/Code-Github-black?logo=github)](https://github.com/yanweifu-sii/CineBrain)
[![Model](https://img.shields.io/badge/Model-CineSync-faa035.svg?logo=Huggingface)](https://huggingface.co/Fudan-fMRI/CineSync)


CineBrain is a large-scale multimodal brain dataset comprising fMRI, EEG, and ECG recordings collected while participants watched episodes of *The Big Bang Theory*.  
It supports research on **neural decoding**, **multimodal learning**, and **modality transfer** in naturalistic narrative processing.

---

## 🧠 Dataset Overview

### Subjects & Stimuli
- **Participants**: 6 subjects  
- **Stimuli**: 30 TV episodes (*The Big Bang Theory*)  
- **Used duration**: first 18 minutes per episode  
- **Total recording time**: 6 hours per subject (~36 hours in total)

### Modalities
- fMRI: TR = 0.8 s
- EEG: 64 channels (the first 64 channels in EEG file), 1000 Hz sampling rate.
- ECG: Recorded synchronously with EEG (the last channel in EEG file)

---

## 📂 Data Organization

### Episode Assignment
- **Subjects 1, 2, 6**: Season 7 (Ep 1–10) + Season 9 (Ep 1–10)
- **Subjects 3, 4, 5**: Season 7 (Ep 1–10) + Season 11 (Ep 1–10)

### Video Segmentation
- **30 episodes** → **8,100 video clips**
- Each episode is segmented into **450 clips**
- **Clip length**: 4 seconds

---

## ⏱️ Temporal Alignment & Segmentation

### fMRI
- Sampled every **0.8 s** (TR = 0.8 s)
- A **4-second video clip** corresponds to **5 fMRI samples**

### EEG
- Original EEG is recorded at **1000 Hz**
- EEG is **directly segmented** from the raw 1000 Hz signal into non-overlapping **0.8 s windows**
- Each **4-second video clip** corresponds to **5 EEG segments**
- No frequency-domain decomposition is applied by default; segmentation is purely temporal

> **Summary**:  
> **1 video clip (4 s) ↔ 5 EEG segments ↔ 5 fMRI samples**

---


### Repository Contents
```
├── videos.tar                          # 8,100 video clips (4s each)
├── sub-00xx/
│   ├── fMRI_raw_data.tar              # Raw fMRI (BIDS format)
│   ├── fMRI_preprocessed_data.tar     # Preprocessed fMRI
│   └── EEG_preprocessed_data.tar      # Preprocessed EEG
└── captions-qwen-2.5-vl-7b.json       # Auto-generated captions
```

### Data Statistics
| Modality | Sampling Rate | Duration/Subject | Total Size |
|----------|---------------|------------------|------------|
| fMRI     | TR = 0.8s     | 6 hours          | ~12 GB     |
| EEG      | 1000 Hz, 64ch | 6 hours          | ~72 GB     |
| Video    | 4s clips      | 9 hours (30 eps) | ~2.6 GB    |

---

## 🔗 Raw Data Links

- **Raw EEG** (.mff format): https://www.modelscope.cn/datasets/Jianxionggao/CineBrain_EEG_Raw
- **Raw fMRI** (BIDS format): https://www.modelscope.cn/datasets/Jianxionggao/CineBrain_fMRI_Raw
- **Preprocessed fMRI** (CIFTI format): https://www.modelscope.cn/datasets/Jianxionggao/CineBrain_fMRI_CIFTI
- **Audio Files** https://www.modelscope.cn/datasets/Jianxionggao/CineBrainVideo

---

## 📌 Important Notes
- **Data Release**: Fully open and downloadable  
- **Cross-Dataset Correspondence**: Subjects **1, 2, 3, 4** in CineBrain map to Subjects **6, 8, 1, 4** in [fMRI-Shape](https://huggingface.co/datasets/Fudan-fMRI/fMRI-Shape) and [fMRI-Objaverse](https://huggingface.co/datasets/Fudan-fMRI/fMRI-Objaverse)  
---


## 📎 Additional Information
- **License**: [Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0)  
- **Languages**: English audiovisual content  

### Citation
If you find our paper useful for your research and applications, please cite using this BibTeX:

```bibtex
@misc{gao2025cinebrain,
  title={CineBrain: A Large-Scale Multi-Modal Brain Dataset During Naturalistic Audiovisual Narrative Processing}, 
  author={Jianxiong Gao and Yichang Liu and Baofeng Yang and Jianfeng Feng and Yanwei Fu},
  year={2025},
  eprint={2503.06940},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2503.06940}, 
}