DDN-Data / Rain1400: Synthetic Single-Image Deraining Dataset (Unofficial Mirror)

Unofficial redistribution of DDN-Data (also widely known as "Rain1400"), the synthetic single-image deraining dataset from Fu et al. (CVPR 2017), packaged for direct use with ClearView's dataset pipeline.

Disclaimer

This repository is not an official release of DDN-Data/Rain1400.

DDN-Data was created by Xueyang Fu, Jiabin Huang, Delu Zeng, Yue Huang, Xinghao Ding, and John Paisley. This repository does not claim ownership of any images or metadata, and attributes the dataset to its original creators. (We describe them as "creators" rather than "copyright holders" deliberately — copyright in academic datasets can rest with an author's university, a funding body, or otherwise, under arrangements we have no way to verify from the outside.)

On licensing — please read before relying on this mirror. We looked for a formal license covering this dataset (project page, associated code repositories, paper supplementary material) and found none — the original distribution is a plain Baidu Cloud / Google Drive link with no accompanying terms of use. This is not unique to this dataset: it's common across this research niche (see our SPA-Data and RealRain-1k mirrors for the two datasets in this collection where we did find explicit licenses, by contrast). In the absence of an explicit license, we have chosen to preserve this dataset for research reproducibility while providing full attribution and making the mirror removable on request — this is a decision we're making under that uncertainty, not a claim that we've established a legal right to redistribute it. Concretely:

• Full attribution and citation to the original creators (below).
• No claim of any rights beyond hosting a copy.
• If the copyright holder, or an authorized representative, requests removal, we will comply promptly.

Separately, another unofficial mirror of this same dataset (jinnovation/rainy-image-dataset) has been publicly available since 2018, suggesting this style of redistribution has existed within the research community for some time — though we recognize that this alone doesn't establish any redistribution right; the absence of an objection is not the same as permission.

If you need certainty about redistribution rights (e.g. for commercial use), contact the dataset's creators or their institution directly rather than relying on this mirror.

A related data point: the raw download includes the authors' own readme.txt (reproduced verbatim as original_readme.txt in this repo), which asks anyone using the dataset to cite their two papers. This indicates the authors expected the dataset to be used in subsequent research — it does not, on its own, constitute an explicit redistribution license.

---

Dataset Description

DDN-Data (Rain1400) is a synthetic single-image deraining dataset: 1,000 clean real photographs, each with 14 synthetically rendered rain streaks overlaid at different orientations and magnitudes (per the authors' own readme.txt, generated using Photoshop's rain-effect technique). It was introduced alongside the "Deep Detail Network" (DDN) architecture and remains one of the most widely used deraining benchmarks — it's also one of the four datasets composited into the Rain13K training set used by MPRNet, Restormer, and most subsequent SOTA deraining papers (there listed as "Rain14000").

• Training: 900 clean images (randomly selected by the authors) × 14 rain variants = 12,600 rainy images.
• Testing: the remaining 100 clean images × 14 rain variants = 1,400 rainy images.

---

Changes from the Official Release

None beyond repackaging for direct download. No images added, removed, or modified. No relabeling. The training/{rainy_image,ground_truth} and testing/{rainy_image,ground_truth} directory layout matches the dataset's conventional distribution structure exactly.

---

Dataset Structure

ddn-data/
├── README.md
├── banner.jpg
├── original_readme.txt
├── training/
│   ├── rainy_image/
│   │   ├── part_1/     # {clean_id}_{variant}.jpg — 6,300 images
│   │   └── part_2/     # {clean_id}_{variant}.jpg — 6,300 images
│   └── ground_truth/   # {clean_id}.jpg — 900 images
└── testing/
    ├── rainy_image/    # {clean_id}_{variant}.jpg — 1,400 images
    └── ground_truth/   # {clean_id}.jpg — 100 images

Note on training/rainy_image's part_1/part_2 split: this is not a meaningful data split (not a train/val split, not grouped by variant or scene) — it exists purely because Hugging Face Hub rejects any single directory containing more than 10,000 files, and the flat 12,600-file rainy_image folder exceeds that. The two subfolders are an arbitrary alternating-file split of the same flat set. ClearView's Rain1400Dataset/DDNDataDataset parser scans rainy_dir recursively, so pointing it at training/rainy_image (the parent of both subfolders) finds all 12,600 files transparently — no manual merging needed.

Each clean image maps to 14 rainy variants via filename pattern {clean_id}_{variant}.jpg → {clean_id}.jpg (e.g. 100_1.jpg through 100_14.jpg → 100.jpg) — this is why ClearView ships a dedicated Rain1400Dataset/DDNDataDataset parser rather than a generic exact-stem matcher.

Split	Clean images	Rainy images (pairs)
training	900	12,600
testing	100	1,400
Total	1,000	14,000

---

Usage with ClearView

This dataset is designed to be used directly with ClearView, an open-source PyTorch framework for image deraining.

from huggingface_hub import snapshot_download
from clearview.data import Rain1400Dataset, get_train_transforms, get_val_transforms

data_dir = snapshot_download(repo_id="dronefreak/DDN-Data", repo_type="dataset")

train_ds = Rain1400Dataset(
    rainy_dir=f"{data_dir}/training/rainy_image",
    clean_dir=f"{data_dir}/training/ground_truth",
    transform=get_train_transforms(crop_size=(256, 256)),
)
test_ds = Rain1400Dataset(
    rainy_dir=f"{data_dir}/testing/rainy_image",
    clean_dir=f"{data_dir}/testing/ground_truth",
    transform=get_val_transforms(),
)

rainy, clean = train_ds[0]

Or directly via the training CLI:

clearview-train \
  --data-dir <path-to-downloaded-snapshot> \
  --dataset-type rain1400 \
  --train-rainy training/rainy_image --train-clean training/ground_truth \
  --val-rainy testing/rainy_image --val-clean testing/ground_truth \
  --model unet --batch-size 16 --crop-size 256 --epochs 100 \
  --loss l1 --mixed-precision \
  --output-dir ./runs/ddn_unet_baseline

Note that evaluate.py applies no resize/crop by default, and this dataset's test images are native (variable) resolution — keep --batch-size 1 for evaluation (its default) to avoid a batch-collation crash.

---

Dataset Sources

The authors report this same dataset across two papers — please cite both if you use it (see Citation).

Original Papers

Removing Rain from Single Images via a Deep Detail Network

Xueyang Fu, Jiabin Huang, Delu Zeng, Yue Huang, Xinghao Ding, John Paisley

IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2017, pp. 1715–1723.

• CVF Open Access (PDF): https://openaccess.thecvf.com/content_cvpr_2017/papers/Fu_Removing_Rain_From_CVPR_2017_paper.pdf
• No arXiv preprint is available for this paper (CVPR proceedings only).

Clearing the Skies: A Deep Network Architecture for Single-Image Rain Removal

Xueyang Fu, Jiabin Huang, Xinghao Ding, Yinghao Liao, John Paisley

IEEE Transactions on Image Processing, vol. 26, no. 6, pp. 2944–2956, 2017.

• arXiv: https://arxiv.org/abs/1609.02087

Official Resources

• Author's project page: https://xueyangfu.github.io/projects/cvpr2017.html (dataset download links: Baidu Cloud, Google Drive)
• Research group homepage: https://xmu-smartdsp.github.io/
• Original readme.txt (bundled with the raw download, reproduced verbatim in this repo as original_readme.txt)

Precedent Unofficial Mirror

• GitHub: https://github.com/jinnovation/rainy-image-dataset — public since 2018, credits the same original authors, no formal license stated there either.

---

Attribution

All credit for the dataset belongs entirely to the original authors: Xueyang Fu, Jiabin Huang, Delu Zeng, Yue Huang, Xinghao Ding, Yinghao Liao, and John Paisley (across the two papers that report this dataset).

If you use this dataset in your research, please cite both original publications below.

---

License

No formal license was located for this dataset — see the Disclaimer above for our reasoning for mirroring it anyway under that uncertainty (full attribution, reproducibility purpose, removable on request). This repository makes no claim to any rights over the dataset content beyond hosting a copy.

---

Citation

If you use this dataset, please cite both papers, per the authors' own request in their original readme.txt:

@InProceedings{Fu_2017_CVPR,
  author = {Fu, Xueyang and Huang, Jiabin and Zeng, Delu and Huang, Yue and Ding, Xinghao and Paisley, John},
  title = {Removing Rain From Single Images via a Deep Detail Network},
  booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  month = {July},
  year = {2017},
  pages = {1715-1723}
}

@article{Fu_2017_TIP,
  author = {Fu, Xueyang and Huang, Jiabin and Ding, Xinghao and Liao, Yinghao and Paisley, John},
  title = {Clearing the Skies: A Deep Network Architecture for Single-Image Rain Removal},
  journal = {IEEE Transactions on Image Processing},
  volume = {26},
  number = {6},
  pages = {2944-2956},
  year = {2017}
}

@article{DBLP:journals/corr/FuHDLP16,
  author       = {Xueyang Fu and
                  Jiabin Huang and
                  Xinghao Ding and
                  Yinghao Liao and
                  John W. Paisley},
  title        = {Clearing the Skies: {A} deep network architecture for single-image
                  rain removal},
  journal      = {CoRR},
  volume       = {abs/1609.02087},
  year         = {2016},
  url          = {http://arxiv.org/abs/1609.02087},
  eprinttype   = {arXiv},
  eprint       = {1609.02087},
  timestamp    = {Sat, 24 Jun 2023 08:49:12 +0200},
  biburl       = {https://dblp.org/rec/journals/corr/FuHDLP16.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}

---

Acknowledgements

We sincerely thank Xueyang Fu, Jiabin Huang, Delu Zeng, Yue Huang, Xinghao Ding, Yinghao Liao, and John Paisley for creating and publicly releasing this foundational deraining benchmark.