2.5K paired sRGB images captured by the main and wide cameras of the Huawei P40 Pro smartphone. The images are perfectly aligned for supervised training and testing.

~500 high-res RGB and synthetic multispectral (MS) images, with ground-truth demosaiced images, designed for real-world dual-camera setup in smartphones.

LLM-generated image and class descriptions for the test sets of the following datasets: ImageNet, CIFAR, Food, SUN, Cars, DTD, Caltech, Pets, and Places.

~1,200 pairs of camera-rendered sRGB and the corresponding scene-referred CIE XYZ images (971 training, 50 validation, and 244 testing images)

A set of paired/unpaired raw images captured by iPhone X and Samsung S9 for raw-to-raw mapping.  DNG and JPG files are available for other research purposes.

~7,000 portrait face images extracted from the WikiArt dataset

~24,000 8-bit sRGB images with different exposure settings and the corresponding well-exposed ground-truth images

 ~4,000 landscape images collected from Flickr with the following licenses: no known copyright restrictions, Public Domain Dedication (CC0), and Public Domain Mark

~100,000 rendered sRGB images with different white-balance (WB) settings; ground truth sRGB images are provided for the training/testing sets

CIFAR-10 testing set of ~15,000 images with realistic in-camera white-balance errors

Over 2,000 high-quality images with their dual-pixel views. It contains scenes with diverse image contents, scene illuminations, and day/night scenes. 

A synthetic set of 150 images to evaluate white-balance methods on mixed-illuminant scenes 

100 images rendered from sensor raw-RGB space to the sRGB (with camera's nonlinear picture style), ProPhoto, Display P3, and Adobe RGB color spaces. 

100 dual-pixel images (left, right, and combined) with corresponding ground truth "depth" maps .

Over 40,000 face images with glasses under different illumination conditions provided with face ROI, facial landmarks and expression, subject ID, gender, and age 

~11,000 hand images (dorsal and palmar sides) of 190 subjects; varying ages; provided with: subject IDs, gender, age, skin color, and more

One week of compressed surveillance videos & two uncompressed surveillance videos

~33,000 labeled answer boxes extracted from ~700 real MCQ answer sheets