45.16 Datasets🔗⧉

Modern imaging is as much about data as about algorithms: a learned denoiser is only as good as the noisy/clean pairs it trained on, and a "state-of-the-art" number means nothing without the benchmark it was measured on (the data story of Chapter 7.2, Machine learning). The benchmark often defines the task: its choices of scene, noise model, and ground truth become the model's assumptions, and its blind spots become the model's blind spots. This appendix is a working index of the public datasets the methods in this book lean on, grouped by what they are for. Each entry is a name, a one-line description, and a link; the chapters that use them point back here rather than reprinting URLs inline.

45.16.1 Classification and features🔗⧉

• ImageNet: the million-image labeled classification set; an influential pretraining corpus for visual recognition. <https://image-net.org>
• COCO: Common Objects in Context: detection, segmentation, keypoints, and captioning, on cluttered everyday scenes. <https://cocodataset.org>
• Places / Places2: scene-recognition at scale (millions of images across hundreds of scene categories); also the standard backdrop for inpainting and scene parsing. <http://places2.csail.mit.edu>

45.16.2 Super-resolution🔗⧉

• DIV2K: 2K-resolution, high-quality images; the de-facto standard super-resolution training set. <https://data.vision.ee.ethz.ch/cvl/DIV2K>
• Flickr2K: 2K Flickr photographs, commonly pooled with DIV2K to enlarge the training pool.
• Set5 / Set14 / BSD100 / Urban100: the small, classic super-resolution test sets (a handful of stock images, natural scenes, and self-similar urban structure) that nearly every SR paper reports on.

45.16.3 Deblurring and restoration🔗⧉

• GoPro (Nah et al. 2017): sharp/blurry video-frame pairs synthesized from high-fps GoPro footage; the standard dynamic-scene motion-deblurring benchmark. <https://seungjunnah.github.io/Datasets/gopro>
• REDS: REalistic and Dynamic Scenes (the NTIRE challenge set): high-quality video for deblurring, super-resolution, and denoising. <https://seungjunnah.github.io/Datasets/reds>

45.16.4 Denoising🔗⧉

• SIDD: Smartphone Image Denoising Dataset: real noisy/clean pairs from phone cameras, the benchmark that pushed denoisers toward realistic sensor noise. <https://www.eecs.yorku.ca/~kamel/sidd/>
• DND: the Darmstadt Noise Dataset: real photographs with held-out ground truth, scored on a submission server to reduce direct tuning to reference images; repeated benchmark-driven selection can still overfit the benchmark. <https://noise.visinf.tu-darmstadt.de>
• Kodak: the 24 lossless "kodim" test images, a long-standing benchmark for denoising, demosaicking, and compression. <https://r0k.us/graphics/kodak/>

45.16.5 HDR and tone mapping🔗⧉

• HDR+ burst dataset: Google's raw burst dataset behind the HDR+ pipeline (Multiple exposure imaging). <https://hdrplusdata.org>
• Kalantari HDR (dynamic scenes): multi-exposure bursts with subject motion, for HDR merging that must handle moving content.
• Laval HDR (sky / indoor): high-dynamic-range outdoor-sky and indoor panoramas, widely used for lighting estimation. <http://hdrdb.com>
• Fairchild HDR Photographic Survey: a set of calibrated HDR scenes built for evaluating tone-mapping operators. <http://markfairchild.org/HDR.html>

45.16.6 Retouching and enhancement🔗⧉

• MIT-Adobe FiveK: 5,000 raw photographs, each retouched by five expert artists; an influential dataset for learned enhancement and tone adjustment. <https://data.csail.mit.edu/graphics/fivek>
• FAID, the Flash and Ambient Illumination Dataset (Aksoy et al. 2018): thousands of registered flash / no-flash image pairs captured in the wild by a crowd; the standard training/benchmark set for learned flash photography (denoising, illumination decomposition, relighting). <https://yaksoy.github.io/flashambient/>

45.16.7 Depth and motion🔗⧉

• NYU Depth V2: indoor RGB-D (Kinect) scenes, the workhorse for monocular depth. <https://cs.nyu.edu/~silberman/datasets/nyu_depth_v2.html>
• KITTI: autonomous-driving benchmark with stereo, depth, and optical-flow ground truth. <https://www.cvlibs.net/datasets/kitti/>
• Middlebury: the classic high-accuracy stereo (and early optical-flow) benchmark. <https://vision.middlebury.edu/stereo/>
• MPI Sintel: a synthetic optical-flow benchmark with long-range, large-motion sequences from an animated film. <http://sintel.is.tue.mpg.de>
• Vimeo-90K: septuplet/triplet clips, the standard training set for frame interpolation and video super-resolution. <http://toflow.csail.mit.edu>

45.16.8 Light fields🔗⧉

• INRIA Lytro Illum Light-Field Dataset (Le Pendu, Guillemot & Dansereau): real plenoptic captures from a Lytro Illum, shipped both as raw .lfr files and decoded 4-D light fields; the source of this book's interactive light-field viewer. The exact decode pipeline (raw lenslet image → uniform 4-D grid) is documented in Chapter 45.23, Reading a Lytro light field. <https://www.irisa.fr/temics/demos/IllumDatasetLF/index.html>
• Stanford Light Field Archive: the classic gantry-captured light fields (incl. the wide-baseline "chess"), used here as the dramatic-refocus reference. <http://lightfield.stanford.edu/lfs.html>

45.16.9 Color and white balance🔗⧉

• NUS (Cheng et al.): a widely used multi-camera color-constancy benchmark: ~1,736 high-bit linear-raw images across 8 cameras (~210 each), each with a ColorChecker in scene for the ground-truth illuminant; its multiple sensors test the cross-camera generalization that actually matters. <https://cvil.eecs.yorku.ca/projects/public_html/illuminant/illuminant.html>
• Gehler-Shi (ColorChecker): the classic baseline: ~568 images from 2 Canon bodies with an in-scene ColorChecker, reprocessed by Shi-Funt; the long-standing reference everyone still reports for comparability. <https://www2.cs.sfu.ca/~color/data/shi_gehler/>
• Cube+: single-illuminant color-constancy images with a calibration cube for ground truth. <https://ipg.fer.hr/ipg/resources/color_constancy>
• X-Rite ColorChecker (24-patch): not a dataset but the reference chart behind most of the above: the 24-patch target used to calibrate color and recover the scene illuminant. <https://www.xrite.com/categories/calibration-profiling/colorchecker-classic>

45.16.10 Faces🔗⧉

• CelebA / CelebA-HQ: celebrity faces with attribute labels; the HQ variant is the high-resolution version used in generative work. <https://mmlab.ie.cuhk.edu.hk/projects/CelebA.html>
• FFHQ: Flickr-Faces-HQ: 70,000 high-quality aligned faces, the StyleGAN training set. <https://github.com/NVlabs/ffhq-dataset>
• LFW: Labeled Faces in the Wild: the long-standing face-verification benchmark. <http://vis-www.cs.umass.edu/lfw/>

45.16.11 Inpainting, segmentation, and matting🔗⧉

• ADE20K: densely annotated scene parsing / semantic segmentation. <https://groups.csail.mit.edu/vision/datasets/ADE20K/>
• Composition-1k: the standard alpha-matting benchmark, foregrounds composited over many backgrounds. <https://sites.google.com/view/deepimagematting>
• DAVIS: Densely-Annotated VIdeo Segmentation: the standard video object segmentation benchmark, with pixel-accurate masks over time. <https://davischallenge.org>

45.16.12 Image quality🔗⧉

• LIVE / TID2013 / KADID-10k: human-rated image-quality databases (distorted images with mean opinion scores) used to validate metrics like PSNR, SSIM, and LPIPS against human perception. <https://live.ece.utexas.edu/research/quality/> · <https://www.ponomarenko.info/tid2013.htm>