README.md

---
license: other
license_name: nvidia-evaluation-data-license
license_link: LICENSE
configs:
- config_name: vqa
  data_files:
  - split: test
    path: data/vqa/data_jsons/annotations/*.json
- config_name: temporal_localization
  data_files:
  - split: test
    path: data/temporal_localization/data_jsons/annotations/*.json
- config_name: event_verification
  data_files:
  - split: test
    path: data/event_verification/filtered/**/*.json
- config_name: referring
  data_files:
  - split: test
    path: data/referring/refdrone_test_llava.json
- config_name: pointing
  data_files:
  - split: test
    path: data/pointing/Vantage2DPointing.tsv
- config_name: tracking
  data_files:
  - split: test
    path: data/tracking/sot_benchmark.jsonl
- config_name: 2dbbox
  data_files:
  - split: test
    path: data/2dbbox/metadata.jsonl
- config_name: dense_captioning
  data_files:
  - split: test
    path: data/dense_captioning/metadata.jsonl
---

# VANTAGE-BENCH

*Video ANalysis Tasks Across Generalized Environments*

## Dataset Description

VANTAGE-BENCH is the first public benchmark purpose-built for evaluating visual understanding on video captured by fixed infrastructure cameras. It spans three real-world domains — warehouse, smart city / Intelligent Transportation Systems (ITS), and smart spaces — across six spatio-temporal video understanding tasks including video question answering (VQA), temporal grounding, dense video captioning, event verification, spatial grounding, and spatio-temporal tracking.

This dataset is for evaluation purposes only.

## Dataset Owner(s)

NVIDIA Corporation

## Dataset Creation Date

April 24, 2026

## License/Terms of Use

This dataset is released under the [NVIDIA Evaluation Data License](./LICENSE.md).

## Dataset Characterization

**Data Collection Method**<br>
Hybrid: Human, Synthetic, Automated. Video data is sourced from vendor-provided footage (GoPro captures of warehouse and smart space environments), synthetic generation (DriveSim collision and multi-camera scenarios), and publicly scraped sources (Dubuque highway/ITS footage).

**Labeling Method**<br>
Hybrid: Human, Synthetic, Pseudolabeled. Annotations for VQA, dense video captions, and temporal localization are primarily human-authored. Spatial grounding labels (2D/3D bounding boxes, referring expressions) use a combination of human annotation and pseudolabeling pipelines (detection + SAM for spatial pointing). Event verification labels are human-curated. Annotations are held server-side for evaluation only.

### Directory Structure

```text
VANTAGE-BENCH/
├── vqa/                     # Video question answering
├── dense_captioning/        # Dense video captioning
├── temporal_localization/   # Temporal localization
├── event_verification/      # Event verification
├── 2dbbox/                  # 2D object localization
├── referring/               # 2D referring expressions
├── pointing/                # 2D spatial pointing
├── tracking/                # Spatio-temporal tracking
└── README.md                # Dataset documentation and submission instructions
```

## Evaluation

### Tasks and Submission Formats

| Category | Task | Metric |
|----------|------|--------|
| Semantic | VQA | Accuracy |
| Semantic | Event Verification | F1 Score |
| Temporal | Dense Video Captioning | SODA-c |
| Temporal | Temporal Localization | mAP@tIoU |
| Spatial | 2D Object Localization | F1@0.5 |
| Spatial | 2D Referring Expressions | mIoU |
| Spatial | 2D Spatial Pointing | Pointing Accuracy |
| Spatio-Temporal | Single Object Tracking | AUC |

Expected submission formats and the leaderboard will be published soon.

### Metric Notes

- **Accuracy**: Percentage of correct predictions.
- **SODA-c**: Metric for dense video captioning quality across event coverage and language quality.
- **mAP@tIoU**: Mean Average Precision measured over temporal IoU thresholds.
- **F1 Score**: Harmonic mean of precision and recall.
- **F1@0.5**: F1 score at an IoU threshold of 0.5.
- **mIoU**: Mean Intersection over Union — average overlap between predicted and ground-truth bounding boxes.
- **Pointing Accuracy**: Percentage of correctly selected target regions.
- **AUC**: Area under the ROC curve, measuring the model's ability to distinguish correct detections or tracks from incorrect ones across varying confidence thresholds.

### Evaluation Server

The [VANTAGE-Bench GitHub repository](https://github.com/anon-benchmark/VANTAGE-bench) provides a sample evaluation pipeline for generating model predictions. Predictions are submitted to the official leaderboard, which will go live by the end of May 2026.

## Dataset Format

Video (mp4) and Images (jpg).

## Dataset Quantification

| Category | Task | Media | Entries |
|----------|------|-------|---------|
| Semantic | VQA | 282 videos | 1,195 QAs (MCQ) |
| Semantic | Event Verification | 163 videos | 163 QAs (BCQ) |
| Temporal | Dense Video Captioning | 104 videos | 717 Events |
| Temporal | Temporal Localization | 203 videos | 1,067 Segments / Spans |
| Spatial | 2D Object Localization | 628 images (3 video sequences) | 27,404 Bboxes |
| Spatial | 2D Referring Expressions | 1,503 images | 3,276 Expressions |
| Spatial | 2D Spatial Pointing | 361 images | 1,005 QAs (MCQ) |
| Spatio-Temporal | Single Object Tracking | 102 video clips | 200 Trajectories |

**Total Entries (Annotations):** 35,027
**Total Media Samples (across tasks, with overlaps):** 3,346
**Total Data Storage:** 42 GB

## Potential Known Risks

- Ground truth annotations are not publicly released. All evaluation is performed server-side.
- Some warehouse videos are concatenated clips from longer recording sessions.

## Citations

```bibtex
@inproceedings{Fujita2020SODA,
  author    = {Soichiro Fujita and Tsutomu Hirao and Hidetaka Kamigaito and Manabu Okumura and Masaaki Nagata},
  title     = {{SODA}: Story Oriented Dense Video Captioning Evaluation Framework},
  booktitle = {Proc. ECCV},
  year      = {2020}
}

@inproceedings{Fu2024BLINK,
  author    = {Xingyu Fu and Yushi Hu and Bangzheng Li and Yu Feng and Haoyu Wang and Xudong Lin and Dan Roth and Noah A. Smith and Wei-Chiu Ma and Ranjay Krishna},
  title     = {{BLINK}: Multimodal Large Language Models Can See but Not Perceive},
  booktitle = {Proc. ECCV},
  year      = {2024}
}

@article{Sun2025RefDrone,
  author    = {Zhichao Sun and Yuda Zou and Xian Sun and Yingchao Feng and Wenhui Diao and Menglong Yan and Kun Fu},
  title     = {{RefDrone}: A Challenging Benchmark for Referring Expression Comprehension in Drone Scenes},
  journal   = {arXiv preprint arXiv:2502.00392},
  year      = {2025}
}
```

## References

- HuggingFace dataset: [nvidia/PhysicalAI-VANTAGE-Bench](https://huggingface.co/datasets/nvidia/PhysicalAI-VANTAGE-Bench)

<img src="./assets/vantage_bench_tasks.png" alt="VANTAGE-BENCH task overview across Semantic, Temporal, Spatial, and Spatio-Temporal understanding categories" width="100%">

## Ethical Considerations

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications.  When downloaded or used in accordance with our terms of service, developers should work with their internal developer teams to ensure this dataset meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
Please report quality, risk, security vulnerabilities or NVIDIA AI Concerns [here](https://app.intigriti.com/programs/nvidia/nvidiavdp/detail).

## Changelog

See [CHANGELOG.md](./CHANGELOG.md) for release history.