README.md

---
license: other
pretty_name: RAD with pixel masks
task_categories:
- image-classification
- image-segmentation
tags:
- anomaly-detection
- industrial-inspection
- 3d
- multi-view
size_categories:
- 1K<n<10K
---

# RAD with pixel masks

Real-world multi-view **A**nomaly **D**etection dataset (RAD), released here with refined
**pixel-level ground-truth masks**.

- Paper: https://arxiv.org/abs/2410.00713
- Project page: https://chang-xinhai.github.io/rad-website/
- Code: https://github.com/kaichen-z/RAD

## Files

| File | Size | Contents |
|---|---|---|
| `Anomaly_refine_msk.zip` | 201 MB | RAD **with pixel masks** — includes `ground_truth/` masks per anomaly type |
| `Anomaly_refine_nonmsk.zip` | 5.85 GB | RAD **without masks** — full-resolution image set, no `ground_truth/` |

## Structure

```text
Anomaly_refine_msk/
├── binderclip/
│   ├── train/good/                 # normal training views
│   ├── test/good/                  # normal test views
│   ├── test/{missing,scratched,…}/ # anomalous test views
│   ├── ground_truth/{missing,…}/   # pixel masks
│   └── transforms.json             # camera metadata
├── binderclip2/
└── …                               # 18 physical-instance directories
```

The paper reports 13 semantic categories: binder clip, bowl, box, can, charger, cup 1, cup 2,
glue bottle, phone case, rubber duck, spoon, spray bottle, tennis ball. Multiple physical
instances of binder clip, cup 2, glue bottle and phone case are kept in separate directories and
aggregated into their semantic category for reporting.

## Download

```bash
pip install -U huggingface_hub

# just the masked split (201 MB)
hf download zhouk777/RAD Anomaly_refine_msk.zip --repo-type dataset --local-dir .
unzip Anomaly_refine_msk.zip

# the full unmasked split (5.85 GB)
hf download zhouk777/RAD Anomaly_refine_nonmsk.zip --repo-type dataset --local-dir .
```

Python:

```python
from huggingface_hub import hf_hub_download

path = hf_hub_download("zhouk777/RAD", "Anomaly_refine_msk.zip", repo_type="dataset")
```

Validate an extracted archive with the release utilities:

```bash
rad-benchmark validate /path/to/Anomaly_refine_msk --require-poses
```

## Citation

```bibtex
@misc{zhou2024rad,
  title         = {RAD: A Realistic Multi-View Benchmark for Pose-Agnostic Anomaly Detection},
  author        = {Zhou, Kaichen and Chang, Xinhai and Kim, Taewhan and Zhang, Jiadong and Cao, Yang and Peng, Chufei and Zhan, Fangneng and Zhao, Hao and Dong, Hao and Ting, Kai Ming and Zhu, Ye},
  year          = {2024},
  eprint        = {2410.00713},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  doi           = {10.48550/arXiv.2410.00713},
  url           = {https://arxiv.org/abs/2410.00713}
}
```

The code license of the GitHub repository does not automatically grant rights to the dataset;
consult the dataset authors for its applicable terms.