README.md

---
license: apache-2.0
task_categories:
- question-answering
- multiple-choice
- visual-question-answering
language:
- en
tags:
- music
pretty_name: AV_Odyssey
size_categories:
- 1K<n<10K
dataset_info:
  features:
  - name: question_id
    dtype: string
  - name: question_type_id
    dtype: int16
  - name: data_type
    dtype: string
  - name: subfield
    dtype: string
  - name: question
    dtype: string
  - name: options
    dtype: string
  - name: answer
    dtype: string
  - name: image_1
    sequence: image
  - name: image_2
    sequence: image
  - name: image_3
    sequence: image
  - name: image_4
    sequence: image
  - name: video_1
    sequence: video
  - name: audio_1
    sequence: audio
  - name: audio_2
    sequence: audio
  - name: audio_3
    sequence: audio
  - name: audio_4
    sequence: audio
  splits:
  - name: test
    num_bytes: 27221062957.18
    num_examples: 4555
  download_size: 27159381702
  dataset_size: 27221062957.18
configs:
- config_name: default
  data_files:
  - split: test
    path: av_odyssey_part*
---

Official dataset for the paper "[AV-Odyssey: Can Your Multimodal LLMs Really Understand Audio-Visual Information?]()".

🌟 For more details, please refer to the project page with data examples: [https://av-odyssey.github.io/](https://av-odyssey.github.io/).

[[🌐 Webpage](https://av-odyssey.github.io/)] [[📖 Paper]()] [[🤗 Huggingface AV-Odyssey Dataset](https://huggingface.co/datasets/AV-Odyssey/AV_Odyssey_Bench)] [[🤗 Huggingface Deaftest Dataset](https://huggingface.co/datasets/AV-Odyssey/Deaftest_dataset)] [[🏆 Leaderboard](https://huggingface.co/spaces/AV-Odyssey/AV_Odyssey_Bench_Leaderboard)]


---

## 🔥 News
* **`2024.11.24`** 🌟 We release AV-Odyssey, the first-ever comprehensive evaluation benchmark to explore whether MLLMs really understand audio-visual information.



## 👀 About AV-Odyssey

Recently, multimodal large language models (MLLMs), such as GPT-4o, Gemini 1.5 Pro, and Reka Core, have expanded their capabilities to include vision and audio modalities. While these models demonstrate impressive performance across a wide range of audio-visual applications, our proposed **DeafTest** reveals that MLLMs often struggle with simple tasks humans find trivial: 1) determining which of two sounds is louder, and 2) determining which of two sounds has a higher pitch. Motivated by these observations, we introduce **AV-Odyssey Bench**. This benchmark encompasses **26** different tasks and **4,555** carefully crafted problems, each incorporating text, visual, and audio components. All data are **newly collected and annotated by humans**, not from any existing audio-visual dataset. AV-Odyssey Bench demonstrates three major features: 1. **Comprehensive** Audio Attributes; 2. **Extensive** Domains; 3. **Interleaved** Text, Audio, and Visual components.

<img src="/assets/intro.png" style="zoom:50%;" />

## 📐 Data Examples

Please refer to our project page https://av-odyssey.github.io/ for exploring more examples.


### 📍AV-Odyssey Bench
<div align="center">
  <img src="assets/demo-1.svg" width="100%" />
</div>


## 🔍 Dataset

**License**:
```
AV-Odyssey is only used for academic research. Commercial use in any form is prohibited.
The copyright of all videos belongs to the video owners.
If there is any infringement in AV-Odyssey, please email [email protected] and we will remove it immediately.
Without prior approval, you cannot distribute, publish, copy, disseminate, or modify AV-Odyssey in whole or in part. 
You must strictly comply with the above restrictions.
```

Please send an email to **[[email protected]](mailto:[email protected])**. 🌟


## 🔮 Evaluation Pipeline





## 🏆 Leaderboard

### Contributing to the AV-Odyssey Leaderboard

🚨 The [Leaderboard](https://huggingface.co/spaces/AV-Odyssey/AV_Odyssey_Bench_Leaderboard) for AV-Odyssey is continuously being updated, welcoming the contribution of your excellent MLLMs!