README.md

---
license: other
license_name: avi-bench-data-use-policy-v1
license_link: https://github.com/FudanCVL/AVI-Bench/blob/main/DATA_USE_POLICY.md

pretty_name: AVI-Bench
viewer: true

tags:
  - audio-visual
  - omni-mllm
  - multimodal-benchmark
  - icml-2026
  - evaluation-only
  - no-training
  - anti-contamination

task_categories:
  - audio-classification
  - image-classification
  - video-classification
  - visual-question-answering
  - audio-text-to-text
  - video-text-to-text
  - image-text-to-text
  - any-to-any

language:
  - en

size_categories:
  - 1K<n<10K

# Per-task configs. AVI-Bench is an evaluation-only benchmark, so every
# config exposes a single "test" split (no train/validation).
configs:
  # ----- Perception -----
  - config_name: AMIC
    data_files:
      - split: test
        path: "levels/perception/AMIC/data.json"
  - config_name: VMIC
    data_files:
      - split: test
        path: "levels/perception/VMIC/data.json"
  - config_name: AVL
    data_files:
      - split: test
        path: "levels/perception/AVL/data.json"
  - config_name: AVM
    data_files:
      - split: test
        path: "levels/perception/AVM/data.json"

  # ----- Understanding -----
  - config_name: VAR
    data_files:
      - split: test
        path: "levels/understand/VAR/data.json"
  - config_name: AVR
    data_files:
      - split: test
        path: "levels/understand/AVR/data.json"
  - config_name: AVC
    data_files:
      - split: test
        path: "levels/understand/AVC/data.json"

  # ----- Reasoning -----
  - config_name: AVH
    data_files:
      - split: test
        path: "levels/reasoning/AVH/data.json"
  - config_name: VAH
    data_files:
      - split: test
        path: "levels/reasoning/VAH/data.json"
  - config_name: AVQA
    data_files:
      - split: test
        path: "levels/reasoning/AVQA/data.json"
  - config_name: AVLG
    data_files:
      - split: test
        path: "levels/reasoning/AVLG/data.json"

  # ----- Primitive Sensation -----
  - config_name: ASQA
    data_files:
      - split: test
        path: "levels/sensation/ASQA/data.json"
  - config_name: VSQA_I
    data_files:
      - split: test
        path: "levels/sensation/VSQA_I/data.json"
  - config_name: VSQA_V
    data_files:
      - split: test
        path: "levels/sensation/VSQA_V/data.json"
  - config_name: AVSQA
    data_files:
      - split: test
        path: "levels/sensation/AVSQA/data.json"

extra_gated_prompt: |
  ## AVI-Bench Data Use Policy v1.0

  AVI-Bench is released **for academic evaluation only**.

  By requesting access you confirm that you have read and accept the
  AVI-Bench Data Use Policy v1.0 in full:

      https://github.com/FudanCVL/AVI-Bench/blob/main/DATA_USE_POLICY.md

  In particular, you agree:

  * **NOT** to use the dataset, in whole or in part, to train,
    fine-tune, distil, align, or otherwise update **any** machine-
    learning model — including LLMs, VLMs, audio-language models,
    omni-modal foundation models, diffusion models, and any subsequent
    model class.
  * **NOT** to construct training data through paraphrasing,
    translation, augmentation, synthetic generation, or LLM-assisted
    relabelling of AVI-Bench content.
  * **NOT** to redistribute, mirror, or rehost the dataset outside this
    Hugging Face repository.
  * **NOT** to crawl or batch-download the dataset by automated means.
  * To use the dataset for evaluation, reproducibility, methodology
    research, and benchmarking (academic or commercial), and not for
    any other purpose.
  * To cite the AVI-Bench paper in any derivative publication.

  Misuse may result in access revocation and, where applicable, legal
  remedy.

extra_gated_fields:
  Full Name: text
  Affiliation: text
  Country: country
  Intended use (one sentence): text
  I have read and accept the AVI-Bench Data Use Policy: checkbox

extra_gated_button_content: I accept the policy and request access
---

> ⚠️  **Evaluation-only dataset.**  AVI-Bench is licensed under the
> [AVI-Bench Data Use Policy v1.0](https://github.com/FudanCVL/AVI-Bench/blob/main/DATA_USE_POLICY.md)
> (CC BY-ND 4.0 + Anti-Training Addendum).
> **Using this dataset, in whole or in part, to train, fine-tune,
> distil, align, or otherwise update any machine-learning model is
> expressly prohibited.**  Bulk redistribution and automated scraping
> are also prohibited.  Commercial evaluation and benchmarking are
> permitted.  See the full policy before downloading.

# AVI-Bench

**AVI-Bench: Toward Human-like Audio-Visual Intelligence of Omni-MLLMs**
(ICML 2026)

A cognitively inspired benchmark for evaluating audio-visual intelligence
of Omni-MLLMs. 5,864 samples, 14 tasks, scored by 9 metrics and
summarised through a four-level AVI taxonomy.

- **Code & docs**: <https://github.com/FudanCVL/AVI-Bench>
- **Project page**: <https://fudancvl.github.io/AVI-Bench/>
- **Data Use Policy**: <https://github.com/FudanCVL/AVI-Bench/blob/main/DATA_USE_POLICY.md>

## Layout

```
levels/
├── perception/        # AMIC, VMIC, AVL, AVM
├── understand/        # VAR, AVR, AVC
├── reasoning/         # AVH, VAH, AVQA, AVLG
└── sensation/         # ASQA, VSQA_I, VSQA_V, AVSQA
```

Each task folder contains a single `data.json` (JSON array of records,
schema: `id`, `task`, `subtask`, `input`, `output`). The HF viewer
exposes each task as a separate config with a single `test` split.

## Loading

```python
from datasets import load_dataset

# Pick a task
ds = load_dataset("FudanCVL/AVIBench", "AVQA", split="test")
print(len(ds), ds.features)  # 469 rows
```

## Tasks and sample counts

| Stage | Tasks | Samples |
|------|------|--------:|
| Perception | AMIC (518), VMIC (521), AVL (205), AVM (250) | 1,494 |
| Understanding | VAR (264), AVR (264), AVC (280) | 808 |
| Reasoning | AVH (250), VAH (250), AVQA (469), AVLG (503) | 1,472 |
| Primitive Sensation | ASQA (502), VSQA_I (620), VSQA_V (580), AVSQA (388) | 2,090 |
| **Total** | **14 tasks** | **5,864** |

## Citation

```bibtex
@inproceedings{wang2026avibench,
  title     = {AVI-Bench: Toward Human-like Audio-Visual Intelligence of Omni-MLLMs},
  author    = {Wang, Yaoting and Zhang, Ziyi and Tu, Wenming and Xu, Shaoxuan and Du, Wenjie and Liang, Cheng and Wang, Weijun and Li, Yuanchao and Li, Guangyao and Fei, Hao and Li, Yuanchun and Ding, Henghui and Liu, Yunxin},
  booktitle = {Proceedings of the 43rd International Conference on Machine Learning (ICML)},
  year      = {2026}
}
```

## Attribution

AVI-Bench bundles content derived from several public sources
(including MusicAVQA, AV-Caps, AVHBench, and others). Each upstream
source retains its own license. The Anti-Training Addendum of the
AVI-Bench Data Use Policy governs the AVI-Bench annotations,
organisation, and processing layered on top.

If you believe any bundled source content infringes your rights,
please open an issue at
<https://github.com/FudanCVL/AVI-Bench/issues> with the subject
**"Removal request"**; we will respond within 14 calendar days.