Image Restoration

BSD is a dataset used frequently for image denoising and super-resolution. Of the subdatasets, BSD100 is aclassical image dataset having 100 test images proposed by Martin et al.. The dataset is composed of a large variety of images ranging from natural images to object-specific such as plants, people, food etc. BSD100 is the testing set of the Berkeley segmentation dataset BSD300. Source: A Deep Journey into Super-resolution: A Survey Image Source: https://www.slideshare.net/jbhuang/single-image-super-resolution-from-transformed-selfexemplars-cvpr-2015

DIV2K is a popular single-image super-resolution dataset which contains 1,000 images with different scenes and is splitted to 800 for training, 100 for validation and 100 for testing. It was collected for NTIRE2017 and NTIRE2018 Super-Resolution Challenges in order to encourage research on image super-resolution with more realistic degradation. This dataset contains low resolution images with different types of degradations. Apart from the standard bicubic downsampling, several types of degradations are considered in synthesizing low resolution images for different tracks of the challenges. Track 2 of NTIRE 2017 contains low resolution images with unknown x4 downscaling. Track 2 and track 4 of NTIRE 2018 correspond to realistic mild ×4 and realistic wild ×4 adverse conditions, respectively. Low-resolution images under realistic mild x4 setting suffer from motion blur, Poisson noise and pixel shifting. Degradations under realistic wild x4 setting are further extended to be of different levels from image to image. Source: Unsupervised Image Super-Resolution with an Indirect Supervised Path

The Urban100 dataset contains 100 images of urban scenes. It commonly used as a test set to evaluate the performance of super-resolution models. Image Source: http://vllab.ucmerced.edu/wlai24/LapSRN/

The Set14 dataset is a dataset consisting of 14 images commonly used for testing performance of Image Super-Resolution models. Image Source: https://www.ece.rice.edu/~wakin/images/

The Set5 dataset is a dataset consisting of 5 images (“baby”, “bird”, “butterfly”, “head”, “woman”) commonly used for testing performance of Image Super-Resolution models. Image Source: http://people.rennes.inria.fr/Aline.Roumy/results/SR_BMVC12.html

The GoPro dataset for deblurring consists of 3,214 blurred images with the size of 1,280×720 that are divided into 2,103 training images and 1,111 test images. The dataset consists of pairs of a realistic blurry image and the corresponding ground truth sharp image that are obtained by a high-speed camera. Source: Down-Scaling with Learned Kernels in Multi-Scale Deep Neural Networksfor Non-Uniform Single Image Deblurring Image Source: Deep Multi-scale Convolutional Neural Network for Dynamic Scene Deblurring

Manga109 has been compiled by the Aizawa Yamasaki Matsui Laboratory, Department of Information and Communication Engineering, the Graduate School of Information Science and Technology, the University of Tokyo. The compilation is intended for use in academic research on the media processing of Japanese manga. Manga109 is composed of 109 manga volumes drawn by professional manga artists in Japan. These manga were commercially made available to the public between the 1970s and 2010s, and encompass a wide range of target readerships and genres (see the table in Explore for further details.) Most of the manga in the compilation are available at the manga library “Manga Library Z” (formerly the “Zeppan Manga Toshokan” library of out-of-print manga). Source: Manga109 Image Source: https://arxiv.org/pdf/1510.04389v1.pdf

Berkeley Segmentation Data Set 500 (BSDS500) is a standard benchmark for contour detection. This dataset is designed for evaluating natural edge detection that includes not only object contours but also object interior boundaries and background boundaries. It includes 500 natural images with carefully annotated boundaries collected from multiple users. The dataset is divided into three parts: 200 for training, 100 for validation and the rest 200 for test. Source: Object Contour Detection with a Fully Convolutional Encoder-Decoder Network Image Source: https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/resources.html

The LOL dataset is composed of 500 low-light and normal-light image pairs and divided into 485 training pairs and 15 testing pairs. The low-light images contain noise produced during the photo capture process. Most of the images are indoor scenes. All the images have a resolution of 400×600. Source: Unsupervised Real-world Low-light Image Enhancement with Decoupled Networks Image Source: https://daooshee.github.io/BMVC2018website/

SIDD is an image denoising dataset containing 30,000 noisy images from 10 scenes under different lighting conditions using five representative smartphone cameras. Ground truth images are provided along with the noisy images. Source: A High-Quality Denoising Dataset for Smartphone Cameras

Color BSD68 dataset for image denoising benchmarks is part of The Berkeley Segmentation Dataset and Benchmark. It is used for measuring image denoising algorithms performance. It contains 68 images. Source: https://github.com/clausmichele/CBSD68-dataset

Raindrop is a set of image pairs, where each pair contains exactly the same background scene, yet one is degraded by raindrops and the other one is free from raindrops. To obtain this, the images are captured through two pieces of exactly the same glass: one sprayed with water, and the other is left clean. The dataset consists of 1,119 pairs of images, with various background scenes and raindrops. They were captured with a Sony A6000 and a Canon EOS 60. Source: Attentive Generative Adversarial Network for Raindrop Removal from a Single Image

The Synthetic Rain Datasets consists of 13,712 clean-rain image pairs gathered from multiple datasets (Rain14000, Rain1800, Rain800, Rain12). With a single trained model, evaluation could be performed on various test sets, including Rain100H, Rain100L, Test100, Test2800, and Test1200. PSNR and SSIM are computed on Y-channel in YCbCr color space.

Consists of 8,422 blurry and sharp image pairs with 65,784 densely annotated FG human bounding boxes. Source: Human-Aware Motion Deblurring

A new large-scale benchmark consisting of both synthetic and real-world hazy images, called REalistic Single Image DEhazing (RESIDE). RESIDE highlights diverse data sources and image contents, and is divided into five subsets, each serving different training or evaluation purposes. Source: RESIDE

DRealSR establishes a Super Resolution (SR) benchmark with diverse real-world degradation processes, mitigating the limitations of conventional simulated image degradation. It has been collected from five DSLR cameras in natural scenes and cover indoor and outdoor scenes avoiding moving objects, e.g., advertising posters, plants, offices, buildings. The training images are cropped into 380×380, 272×272 and 192×192 patches, resulting in 31,970 patches. Source: Component Divide-and-Conquer for Real-World Image Super-Resolution

We released a large-scale underwater image (LSUI) dataset including 5004 image pairs, which involve richer underwater scenes (lighting conditions, water types and target categories) and better visual quality reference images than the existing ones.

The PIRM dataset consists of 200 images, which are divided into two equal sets for validation and testing. These images cover diverse contents, including people, objects, environments, flora, natural scenery, etc. Images vary in size, and are typically ~300K pixels in resolution.

Multispectral imaging using multiplexed illumination.

High-quality hyperspectral reconstruction using a spectral prior

Using the validation set (100 images) from the widely used DIV2K dataset, we blurred and subsampled each image with a different, randomly generated kernel. Kernels were 11x11 anisotropic gaussians with random lengths λ1, λ2∼U(0.6, 5) independently distributed for each axis, rotated by a random angle θ∼U[−π, π].

TinyPerson is a benchmark for tiny object detection in a long distance and with massive backgrounds. The images in TinyPerson are collected from the Internet. First, videos with a high resolution are collected from different websites. Second, images from the video are sampled every 50 frames. Then images with a certain repetition (homogeneity) are deleted, and the resulting images are annotated with 72,651 objects with bounding boxes by hand. Source: https://arxiv.org/abs/1912.10664 Image Source: https://arxiv.org/pdf/1912.10664.pdf

Various documents dataset. Each of the 65 documents includes scanned ground truth images, both hard and easy distorted photos, and document-centered cropped images. Image Source: DocUNet: Document Image Unwarping via a Stacked U-Net

An image restoration dataset

The LIVE Public-Domain Subjective Image Quality Database is a resource developed by the Laboratory for Image and Video Engineering at the University of Texas at Austin. It contains a set of images and videos whose quality has been ranked by human subjects. This database is used in Quality Assessment (QA) research, which aims to make quality predictions that align with the subjective opinions of human observers. The database was created through an extensive experiment conducted in collaboration with the Department of Psychology at the University of Texas at Austin. The experiment involved obtaining scores from human subjects for many images distorted with different distortion types. The QA algorithm may be trained on part of this data set, and tested on the rest. The database is available to the research community free of charge. If you use these images in your research, the creators kindly ask that you reference their website and their papers. There are two releases of the database. Release 2 includes more distortion types and more subjects than Release 1. The distortions include JPEG-compressed images, JPEG2000-compressed images, Gaussian blur, and white noise.

The dataset used for NTIRE 2022 Spectral Recovery Challenge

Flare7K, the first nighttime flare removal dataset, which is generated based on the observation and statistic of real-world nighttime lens flares. It offers 5,000 scattering flare images and 2,000 reflective flare images, consisting of 25 types of scattering flares and 10 types of reflective flares. The 7,000 flare patterns can be randomly added to the flare-free images, forming the flare-corrupted and flare-free image pairs.

End-to-End Low Cost Compressive Spectral Imaging with Spatial-Spectral Self-Attention

Five classic grayscale images commonly used for image quality assessment tasks.

A large-scale dataset of ~29.5K rain/rain-free image pairs that covers a wide range of natural rain scenes. Source: Spatial Attentive Single-Image Deraining with a High Quality Real Rain Dataset

The dataset contains single-shot videos taken from moving cameras in underwater environments. The first shard of a new Marine Video Kit dataset is presented to serve for video retrieval and other computer vision challenges. In addition to basic meta-data statistics, we present several insights based on low-level features as well as semantic annotations of selected keyframes. 1379 videos with a length from 2 s to 4.95 min, with the mean and median duration of each video is 29.9 s, and 25.4 s, respectively. We capture data from 11 different regions and countries during the time from 2011 to 2022.

Comprises 4 different subsets - Flat, House, Priory and Lab - each containing a number of different sequences that can be successfully relocalised against each other. Source: CSD

A carefully chosen set of high-resolution high-precision natural images suited for compression algorithm evaluation. The images historically used for compression research (lena, barbra, pepper etc...) have outlived their useful life and its about time they become a part of history only. They are too small, come from data sources too old and are available in only 8-bit precision. These high-resolution high-precision images have been carefully selected to aid in image compression research and algorithm evaluation. These are photographic images chosen to come from a wide variety of sources and each one picked to stress different aspects of algorithms. Images are available in 8-bit, 16-bit and 16-bit linear variations, RGB and gray. These Images are available without any prohibitive copyright restrictions. These images are (c) there respective owners. You are granted full redistribution and publication rights on these images provided: 1. The origin of the pictures must not be misrepresented; you must not claim that you took the original pictures. If you use, publish or redistribute them, an acknowledgment would be appreciated but is not required. 2. Altered versions must be plainly marked as such, and must not be misinterpreted as being the originals. 3. No payment is required for distribution of this material, it must be available freely under the conditions stated here. That is, it is prohibited to sell the material. 4. This notice may not be removed or altered from any distribution. For grayscale evaluation, use the Grayscale 8 bit dataset, for color evaluation, use the Color 8 bit dataset. @online{icb, author = {Rawzor}, title = {Image Compression Benchmark}, url = {http://imagecompression.info/} }

The first ultra-high-definition image demoireing dataset, consisting of 4,500 4K resolution training pairs and 500 standard 4K resolution validation pairs.

The existing multi-modality image fusion dataset lacks comprehensive coverage of adverse weather scenarios. To address this, we introduce AWMM-100k, a benchmark dataset constructed by selecting samples from RoadScene, MSRS, M3FD, and LLVIP, followed by controlled degradation processing to simulate adverse weather conditions. Combined with real-world data captured using a DJI M30T drone equipped with high-resolution visible and thermal cameras, AWMM-100k comprises 187,699 images covering rain, haze, and snow, each categorized into heavy, medium, and light intensities. This dataset supports research on multi-modality image fusion under challenging weather conditions and is also applicable to image restoration tasks such as dehazing, deraining, and desnowing. We thank the original dataset for its contribution. In addition, we believe this dataset significantly expands the scope of multimodal image processing and computer vision research, facilitating advancements in both image fusion and restoration tasks under adverse conditions.