Datasets:

DanielNguyen121
/

Noisy_datasets

image_id int64 43.9k 2.27M
43,890
61,136
61,401
61,581
62,125
62,134
62,672
62,719
62,765
62,769
63,055
63,059
63,096
63,105
63,284
63,293
63,295
63,838
64,005
64,072
64,200
64,793
64,919
64,924
64,932
65,207
65,294
65,328
65,387
65,405
65,406
65,780
65,838
66,278
66,281
66,405
66,406
66,410
66,418
66,420
66,423
66,521
66,530
66,532
66,977
67,031
67,035
67,037
67,618
68,810
68,811
70,035
71,754
72,369
72,491
72,617
74,246
74,462
75,752
76,795
76,843
77,517
78,141
78,400
78,427
82,497
82,500
83,150
85,153
87,248
89,346
91,257
91,354
91,357
91,361
91,403
92,378
92,971
93,510
94,396
95,325
95,379
97,107
97,111
97,578
97,765
100,263
102,560
103,329
103,331
104,601
104,604
104,832
104,855
108,172
109,449
109,458
109,630
109,655
114,397

End of preview. Expand in Data Studio

Noisy-labels project data and provenance

This repository holds research metadata and lawfully redistributable annotation files. The canonical execution source, reports and checkpoints are in DanielNguyen121/Noisy-labels-project.

FGVC-Aircraft 2013b

Official source: https://www.robots.ox.ac.uk/~vgg/data/fgvc-aircraft/ . Photographer copyright and non-commercial research terms apply. No Aircraft JPEGs are mirrored here. Use the pinned project reconstruction script and official archive. The source page's published MD5 disagrees with currently served official bytes; project provenance records the verified current SHA256. metadata/aircraft/ contains hierarchy, split and subset manifests.

CIFAR-100N

Human annotation source: https://github.com/UCSC-REAL/cifar-10-100n at commit 49df7d8a69e355470c77c1c2f2424916325a394b. Annotation files are redistributed under the source Creative Commons Attribution-NonCommercial 4.0 terms with attribution to Wei et al., Learning with Noisy Labels Revisited: A Study Using Real-World Human Annotations, ICLR 2022. The original CIFAR-100 image archive is not mirrored here; obtain it from https://www.cs.toronto.edu/~kriz/cifar.html and verify its published MD5 eb9058c3a382ffc7106e4002c42a8d85.

Phase-B synthetic noise

noise/phaseB/*_observed.csv contains learner-visible observed labels for the frozen seven-family/fourteen-variant Aircraft subset. Evaluator-only clean labels, corruption flags, and oracle ancestors are stored separately in the private project repository; they must not enter non-oracle training.

Downloads last month: 162