RainDS: Combined Rain-Streak and Raindrop Removal Benchmark (Unofficial Mirror)

Unofficial redistribution of RainDS, the combined rain-streak and raindrop removal dataset from Quan et al. (CVPR 2021), packaged for direct use with ClearView's dataset pipeline. Each clean image has up to three separate degraded variants, raindrop-only, rain-streak-only, and both combined, in one dataset.

Disclaimer

This repository is not an official release of RainDS.

RainDS was created by Ruijie Quan, Xin Yu, Yuanzhi Liang, and Yi Yang. 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 (official/community code repository, paper supplementary material) and found none, the original distribution is a plain Google Drive link with no accompanying terms of use. This is the same situation as this collection's Rain13K, DDN-Data, RainDrop, and NTIRE-Haze mirrors, not a stated restriction like a dataset that explicitly says "research purposes only" (which we treat differently and do not mirror, see this collection's decision on GT-RAIN). 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.

A note on image content. RainDS_real's scenes are genuine street photography (parked cars, building facades), and like most real-world driving/street datasets, some frames incidentally include visible license plates. We have not attempted to redact or filter these, consistent with "no changes from the official release" below; if you need a plate-free subset for your use case, screen the images yourself before use.

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.

---

Dataset Overview

RainDS is two collections in one: a synthetic set (RainDS_syn) and a real-world set (RainDS_real), each further split by degradation type. For every clean (gt) image, there can be up to three degraded counterparts:

• raindrop: lens-adherent droplets only (the same physical phenomenon as this collection's RainDrop mirror).
• rainstreak: falling rain streaks only (the same phenomenon as this collection's synthetic/real deraining mirrors).
• rainstreak_raindrop: both combined in the same frame, the actual joint task RainDS was built for.

Collection	Split	Pairs per degradation	Total (×3 degradations)
RainDS_syn	train	1,000	3,000
RainDS_syn	test	200	600
RainDS_real	train	150	450
RainDS_real	test	98 (raindrop, rainstreak_raindrop), 97 (rainstreak)	293
Total			4,343

RainDS_real's test_set/rainstreak folder has one fewer pair than the other two degradation categories in that split, a single file (IMG_7435.png) in the official release has no corresponding numeric ID in gt/, see Changes from the Official Release below.

---

Changes from the Official Release

No images added, removed, or modified, and no files renamed. __MACOSX/ and .DS_Store artifacts left behind by the original zip's macOS packaging were dropped (pure archive noise, not data).

One thing we did not change, flagged rather than silently worked around: RainDS_real/test_set/rainstreak/ ships a file named IMG_7435.png with no matching numeric ID in that split's gt/ folder, an inconsistency in the upstream release itself. ClearView's RainDSDataset parser (below) handles this by matching on the intersection of available IDs and logging a warning, rather than erroring or silently renaming the file to paper over it.

---

Dataset Structure

rainds/
├── README.md
├── banner.jpg
├── RainDS_syn/
│   ├── train/
│   │   ├── gt/                    # norain-{id}.png (or pie-norain-{id}.png) — 1,000 images
│   │   ├── raindrop/               # rd-{id}.png (or pie-rd-{id}.png) — 1,000 images
│   │   ├── rainstreak/             # rain-{id}.png (or pie-rain-{id}.png) — 1,000 images
│   │   └── rainstreak_raindrop/    # rd-rain-{id}.png (or pie-rd-rain-{id}.png) — 1,000 images
│   └── test/                       # same layout, 200 images per category
└── RainDS_real/
    ├── train_set/
    │   ├── gt/                     # {id}.png — 150 images
    │   ├── raindrop/               # {id}.png — 150 images
    │   ├── rainstreak/             # {id}.png — 150 images
    │   └── rainstreak_raindrop/    # {id}.png — 150 images
    └── test_set/                   # same layout, 98 images per category (97 for rainstreak)

RainDS_syn marks image type with a filename prefix (norain-/rd-/rain-/rd-rain-), optionally itself prefixed with pie- for a second sub-collection folded into the same folders with otherwise-overlapping numeric IDs. RainDS_real uses plain identical-stem filenames across all four categories. This is why ClearView ships a dedicated RainDSDataset parser rather than the generic ImagePairDataset: it normalizes both conventions, and specifically preserves the pie- tag as part of the matching ID so the two RainDS_syn sub-collections don't collide with each other.

---

Usage with ClearView

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

from huggingface_hub import snapshot_download
from clearview.data import RainDSDataset, get_train_transforms

data_dir = snapshot_download(repo_id="dronefreak/RainDS", repo_type="dataset")

# Joint raindrop + rain-streak removal (the dataset's primary task):
train_ds = RainDSDataset(
    f"{data_dir}/RainDS_syn/train",
    degradation="rainstreak_raindrop",
    transform=get_train_transforms(crop_size=(256, 256)),
)

# Or isolate one degradation type:
raindrop_only = RainDSDataset(f"{data_dir}/RainDS_real/test_set", degradation="raindrop")
rainstreak_only = RainDSDataset(f"{data_dir}/RainDS_syn/test", degradation="rainstreak")

degraded, clean = train_ds[0]

Note: RainDSDataset is not yet wired into the clearview-train CLI's --dataset-type choices (Python API only for now).

---

Dataset Sources

Original Paper

Removing Raindrops and Rain Streaks in One Go

Ruijie Quan, Xin Yu, Yuanzhi Liang, Yi Yang

IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021.

• CVPR Open Access (PDF): https://openaccess.thecvf.com/content/CVPR2021/papers/Quan_Removing_Raindrops_and_Rain_Streaks_in_One_Go_CVPR_2021_paper.pdf

Official Resources

• Code + Dataset: https://github.com/Songforrr/RainDS_CCN

---

Attribution

All credit for the dataset belongs entirely to the original authors: Ruijie Quan, Xin Yu, Yuanzhi Liang, and Yi Yang.

If you use this dataset in your research, please cite the original publication 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:

@InProceedings{Quan_2021_CVPR,
  author = {Quan, Ruijie and Yu, Xin and Liang, Yuanzhi and Yang, Yi},
  title = {Removing Raindrops and Rain Streaks in One Go},
  booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year = {2021}
}

---

Acknowledgements

We sincerely thank Ruijie Quan, Xin Yu, Yuanzhi Liang, and Yi Yang for creating and publicly releasing this dataset.