You need to agree to share your Hugging Face information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

This collection re-hosts processed versions of public vision datasets for research on generative vision foundation models. Each subset keeps its ORIGINAL license (table below); by requesting access you agree to follow the license and terms of every subset you use and to cite the original authors. Access is granted manually.

Log in or Sign Up to review the conditions and access this dataset content.

pixel-foundation data

Processed training / evaluation data of the pixel-foundation project: vision tasks (perception, low-level, 3D) whose outputs are encoded as RGB images and learned by image-generation / editing models. Each subset = one source dataset, in a unified record format with the raw ground truth (boxes, masks, depth in meters, normals, flow, disparity, camera, ...). The RGB encoding of targets is done online at training time; the preview column shows one such encoding for browsing only.

Columns: uid, dataset, split, tasks, record (JSON; paths relative to the data root), image_path + image (first input image, original bytes), preview, files (every other referenced file — labels, extra views — as {path, bytes}, each file stored once per subset). Writing image / files bytes back to <data_root>/<path> and the records to processed/<dataset>/<split>.jsonl restores the on-disk layout.

subset tasks rows original license
allweather weather_removal train 18,069 unspecified (composite)
ambientcg_hdri camera_calib train 281, val 19 CC0
autoflow optical_flow train 39,610, val 390 unverified (Google research release)
biped edge test 50, train 200 non-commercial research (MIT code)
blendedmvs pointmap_pose, depth train 17,010, val 922 CC BY 4.0
bsd100 super_resolution bicubic_x2 100, bicubic_x3 100, bicubic_x4 100, test 100 research
bsds500 edge test 200, train 200, val 100 non-commercial research
cbsd68 denoise test 68 BSDS non-commercial research
coco detection, instance_seg, pose_2d, semantic_seg train 118,287, val 5,000 CC BY 4.0 (annotations); images Flickr terms
div2k super_resolution, denoise bicubic_x2 100, bicubic_x3 100, bicubic_x4 100, train 800, val 100 academic research only
dut_omron salient_object test 5,168 research use
duts salient_object test 5,019, train 10,553 research use
ecssd salient_object test 1,000 research use
entityseg entity_seg train 31,913, val 1,314
eth3d_two_view stereo train 27 CC BY-NC-SA 4.0
flickr2k super_resolution, denoise train 2,650 Flickr images, research use
flyingchairs optical_flow train 22,232, val 640 research-only, non-commercial
geocalib_openpano camera_calib test 40, train 670, val 39 PolyHaven CC0 + HDRMaps terms
gopro deblur test 1,111, train 2,103 CC BY 4.0
grefcoco referring_seg testA 750, testB 750, train 16,994, val 1,501 CC BY-NC-SA 4.0 (unverified)
hed_bsds_aug edge train 28,800 non-commercial research
him_2k matting test 320, train 1,680 CC BY-NC 4.0
holicity camera_calib test 305, train 5,410, val 297 non-commercial research (Google Street View imagery)
hrsod salient_object test 400, train 1,610 CC BY-NC 4.0 (per mirror card)
hypersim depth, normal, semantic_seg, camera_calib, pointmap_pose test 7,759, train 60,123, val 7,460 CC BY-SA 3.0
kenburns3d normal train 74,035, val 2,013 CC BY-NC-SA 4.0
kitti_2012 stereo, optical_flow train 388 CC BY-NC-SA 3.0
kitti_2015 stereo, optical_flow train 400 CC BY-NC-SA 3.0
kodak24 denoise test 24 Kodak free use
lol_v1 low_light test 15, train 485 unspecified (research)
lol_v2 low_light test 200, train 1,589 unspecified (research)
lsd_lowlight low_light train 63,040, val 1,278 CC BY-NC 4.0
lsdir super_resolution, denoise train 84,991, val 250 LSDIR terms (official repo gated; code MIT)
lvis_v1 detection, instance_seg train 100,170, val 19,809 CC BY 4.0 (annotations); COCO images Flickr terms
megadepth_1500 pointmap_pose test 1,500 research
middlebury_2005 stereo train 54 no explicit license (cite)
middlebury_2006 stereo train 189 no explicit license (cite)
middlebury_2021 stereo train 24 no explicit license (cite)
middlebury_eval3 stereo train 15 no explicit license (cite)
mpii pose_2d train 16,879, val 529 Simplified BSD
msra10k salient_object train 10,000 research use
nyud_edge edge test 654, train 795 research (NYUv2 terms)
nyuv2 depth test 654, train 795 no explicit license (research)
omnidata_replica normal train 98,654, val 5,743 Replica research-only license
omnidata_replica_gso normal train 80,353, val 9,307 Replica research-only license
ostv2 super_resolution, denoise train 10,324 research
polyhaven_hdri camera_calib train 354, val 24 CC0
polyu_noisy denoise train 40 research (repo License.txt)
rain13k weather_removal test_rain100h 100, test_rain100l 100, test_test100 98, test_test1200 1,200, test_test2800 2,800, train 13,711 unknown (composite of Rain14000/Rain1800/Rain800/Rain12)
raindrop weather_removal test_a 58, test_b 249, train 861 unknown
realblur deblur test 1,960, train 7,516 CC BY 4.0
realsr_v3 super_resolution test 300, train 1,265 unspecified (academic)
reds deblur train 4,800, val 600 CC BY 4.0
refcoco referring_seg testA 750, testB 750, train 16,994, val 1,500
refcocog referring_seg test 2,600, train 21,899, val 1,300
refcocoplus referring_seg testA 750, testB 750, train 16,992, val 1,500
rellisur low_light test 425, train 3,610 CC BY 4.0 (original; mirror card says Apache-2.0)
reside_its weather_removal train 13,990 unspecified
reside_sots weather_removal test_indoor 500, test_outdoor 500 unspecified
rsblur deblur test 3,360, train 8,878, val 1,120 unspecified (academic)
sbd edge train 8,498, val 2,857 research
sceneflow_driving stereo train 4,400 research-only, non-commercial
sceneflow_flyingthings3d stereo test 4,368, train 22,383 research-only, non-commercial
sceneflow_monkaa stereo train 8,657 research-only, non-commercial
set14 super_resolution bicubic_x2 14, bicubic_x3 14, bicubic_x4 14, test 14 research
set5 super_resolution bicubic_x2 5, bicubic_x3 5, bicubic_x4 5, test 5 research
sice low_light test 149, train 768 unspecified (research)
sidd_medium denoise train 320 MIT (SIDD)
sidd_val denoise test 1,280 MIT (SIDD)
sintel optical_flow train 2,082 research use (Sintel terms)
spring stereo, optical_flow train 9,773 CC BY 4.0
tartanair depth, optical_flow, stereo train 29,216, val 1,760 BSD-3 (HF card) / CC BY 4.0 (site)
thinobject_5k salient_object test 500, train 5,248 CC BY-NC 4.0
transparent_460 matting test 100, train 4,100 non-commercial research (README.txt)
uhrsd salient_object test 988, train 4,856 CC BY-NC 4.0 (per mirror card)
unrealstereo4k stereo train 8,000, val 200 MIT (HF card)
urban100 super_resolution bicubic_x2 100, bicubic_x3 100, bicubic_x4 100, test 100 research
virtual_kitti2 depth, optical_flow, stereo, camera_calib, pointmap_pose train 25,744, val 16,731 CC BY-NC-SA 3.0
wed super_resolution, denoise train 4,744 research

Datasets whose terms forbid redistribution are not included.

Downloads last month
19,607