RENOIR - A Dataset of Real Low-Light Images

We introduce the first publicly available dataset of images corrupted by real low-light noise together with pixel and intensity aligned clean images. It contains about 500 images of 120 scenes that have been collected in low-light setting using three cameras: Cannon T3i, Cannon S90 and a Xiaomi MI3 mobile phone. The dataset is quite large since the images have the original sensor resolution, so each image has about 10 megapixels.

The dataset could be used by researchers to investigate noise formation and noise statistics in low-light digital camera images, to train and test image denoising algorithms, or other uses. The dataset is challenging since the images have a large range of levels of noise, so a denoising algorithm needs to adjust its internal parameters accordingly for each image.

The process for collecting the images and for aligning their brightness is presented in the following paper:

1. J. Anaya, A. Barbu. RENOIR - A Dataset for Real Low-Light Image Noise Reduction. Journal of Visual Comm. and Image Rep. 51, No. 2, 144-154, 2018 (arxiv)

The images collected with each camera are provided below:

1. Canon T3i    aligned (4.9Gb), raw (4.5Gb)
2. Canon S90   aligned (2.6Gb), raw (1.8Gb)
3. Xiaomi Mi3  aligned (2.1Gb), raw (2.5Gb)

Also provided above are the original 12-bit RAW format images that were collected, which are not brightness-aligned. However, for most applications we recommend using the aligned images.

In the paper is also presented a method for computing the PSNR of one of the images even though none of the images is noise free. A demo of this PSNR calculation is presented here.

If you use this data, please cite the following paper:

1. J. Anaya, A. Barbu. RENOIR - A Dataset for Real Low-Light Image Noise Reduction. Journal of Visual Comm. and Image Rep. 51, No. 2, 144-154, 2018 (arxiv)
Bibtex entry:
@article{anaya2018renoir,
title={RENOIR - A Dataset for Real Low-Light Noise Image Reduction},
author={Anaya, Josue and Barbu, Adrian},
journal={Journal of Visual Communication and Image Representation},
volume={51},
pages={144--154},
year={2018}
}