MhdMohammadi/SIDD-CC-DNG

SIDD-CC — Source DNG Bursts (train split)

This repository hosts the raw DNG burst captures that SIDD-CC is built from, for the train split only: 115 capture instances × 150 DNG frames = 17,250 DNG files, 234.5 GB.

SIDD-CC ("Color-Corrected") is a re-processing of the Smartphone Image Denoising Dataset (SIDD) (Abdelhamed, Lin & Brown, CVPR 2018). The DNG files here are the unmodified originals released by the SIDD authors — no pixels have been changed. What changes in SIDD-CC is how the ground-truth image is estimated from them, and which captures are considered usable.

The train/test split

Part of the original SIDD test set is no longer available from the official distribution, so a new split was needed. Of the 200 SIDD instances, 166 are still served; after excluding Samsung S6 captures whose RAW data was already clipped at black on the device, 144 remain valid, split 115 train / 29 test (~20% test). This repository carries the train half; the 29 test instances are released separately.

Contents

data/train/{instance}_DNG.zip   115 zips, one per capture instance
                                each contains {instance}_DNG/{instance}_DSC_*.DNG  (150 frames)
metadata/train_metadata.csv     instance, scene, camera, ISO, shutter, illuminant, brightness, size

Scene names follow the SIDD convention:

0001_001_S6_00100_00060_3200_L
  |    |   |    |     |    |  |
  |    |   |    |     |    |  brightness code (L / N / H)
  |    |   |    |     |    illuminant temperature (K)
  |    |   |    |     shutter speed
  |    |   |    ISO
  |    |   camera
  |    scene number
  instance id

Train split composition

Instances	115
DNG frames	17,250 (150 per instance)
Scene numbers	10
Cameras	iPhone 7 (49), Google Pixel (26), Motorola Nexus 6 (17), Samsung S6 Edge (13), LG G4 (10)
ISO	50 – 10,000
Illuminant	3200 K, 4400 K, 5500 K
Brightness	normal 65, low 48, high 2
Size	234.5 GB

Usage

from huggingface_hub import hf_hub_download
import zipfile

path = hf_hub_download(
    repo_id="MhdMohammadi/SIDD-CC-DNG",
    filename="data/train/0001_DNG.zip",
    repo_type="dataset",
)
with zipfile.ZipFile(path) as z:
    z.extractall("0001")   # -> 0001/0001_DNG/0001_DSC_*.DNG   (150 frames)

Download the full train split:

hf download MhdMohammadi/SIDD-CC-DNG --repo-type dataset --local-dir SIDD-CC-DNG

Or a single camera, via the metadata:

import pandas as pd
from huggingface_hub import hf_hub_download

meta = pd.read_csv(hf_hub_download("MhdMohammadi/SIDD-CC-DNG",
                                   "metadata/train_metadata.csv", repo_type="dataset"))
for f in meta[meta.camera == "GP"].zip_file:
    hf_hub_download("MhdMohammadi/SIDD-CC-DNG", f, repo_type="dataset", local_dir=".")

Licensing and attribution

The DNG files are redistributed from the original SIDD release and remain subject to the SIDD terms of use. If you use this data, please cite both SIDD-CC and the original dataset:

@article{mohammadi2026low,
  title={Why Low-Light Cameras Go Color Blind: Removing Color Bias in Raw Denoising},
  author={Mohammadi, Mohammad and Honari, Sina and Tsogkas, Stavros and Aumentado-Armstrong, Tristan and Brown, Michael S and Mohomed, Iqbal and Derpanis, Konstantinos G and Levinshtein, Alex and Gilitschenski, Igor},
  journal={arXiv preprint arXiv:2607.11090},
  year={2026}
}

@inproceedings{abdelhamed2018high,
  title     = {A High-Quality Denoising Dataset for Smartphone Cameras},
  author    = {Abdelhamed, Abdelrahman and Lin, Stephen and Brown, Michael S.},
  booktitle = {IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2018}
}