Computer Vision

Image datasets for perception models that have to pass a review.

Object detection, segmentation, classification, pose, OCR. COCO, YOLO, Pascal VOC, KITTI, LabelMe formats — every dataset ships with an Ed25519-signed LQS cert and contamination screening against public benchmarks.

Formats
6
COCO · YOLO · VOC · KITTI · LabelMe · Parquet
Tasks
8+
Detection · segmentation · classification · pose · depth · OCR
Seller payout
85%
Signed sale · no subscription · 15% platform fee
Compliance
Art. 10
EU AI Act · SOC 2 scope · subgroup-equity dim
Compliance fields mapped into every cert
SOC 2
Type II · audit
EU AI Act
Art. 10 aligned
GDPR
Art. 28 + SCCs
NIST AI RMF
MEASURE 2.2
Ed25519
signed cert
Featured datasets

Vision datasets with verifiable quality.

Live listings from the marketplace filtered to computer-vision category. Every card shows signed LQS score, contamination-clean flag, and originality signal.

Tasks covered

Every CV task the review team will ask about.

Object detection
Bounding-box datasets compatible with YOLOv5/v8/v11, Detectron2, and MMDetection training pipelines. Per-class counts + IoU-threshold metadata on every cert.
formats · COCO · YOLO · VOC
Semantic + instance segmentation
Pixel-level masks for scene understanding, medical imaging, satellite analysis. Polygon and RLE mask support validated on upload.
formats · COCO RLE · LabelMe
Image classification
Single- and multi-label datasets organized by class. Drop in as training sets for ResNet, EfficientNet, ViT, ConvNeXt, DINOv2.
splits · train/val/test validated
Pose + keypoint estimation
Keypoint-annotated human and animal imagery for body pose, hand tracking, gesture recognition. Joint confidence + occlusion flags preserved.
schema · COCO-17 · MPII · H36M
Scene + depth understanding
Multi-class street, indoor, aerial scenes for autonomous systems and robotics. Depth maps + optical flow fields where available.
scenes · urban · indoor · aerial
OCR + document AI
Text-in-image datasets with word- and character-level bounding boxes. Page-level layout + reading-order annotations for doc AI pipelines.
tasks · detection + recognition
COCO · JSON YOLO · TXT Pascal VOC · XML KITTI · 3D LabelMe · polygon JSON Parquet · + image refs
What LabelSets adds

Cert fields designed for your vendor-review file.

LabelSets is not another bounding-box exchange. Every published CV dataset carries a signed quality cert that maps directly into the paperwork your risk team already files.

Contamination-clean
Every CV dataset is hashed against COCO, ImageNet, OpenImages, CIFAR splits. Overlap flagged at the cert level so your eval numbers hold up.
against · 40+ public evals
Subgroup equity (Art. 10)
Per-subgroup accuracy breakdowns baked into the cert for EU AI Act Article 10 data-governance filings. Demographic + class-balance metrics.
field · subgroup_equity
Ed25519-signed provenance
Every cert carries a public-key signature + fingerprint. Buyers verify at /verify any time. Revocation registry handles post-facto PII flags.
fingerprint · aa4c070af907e2ea
FAQ

Questions procurement actually asks.

COCO JSON, YOLO TXT, Pascal VOC XML, KITTI 3D, LabelMe polygon JSON, and Parquet with image refs. Every listing shows its format. The marketplace filter rail lets you narrow by format.
Every upload runs through the LQS v3.1 pipeline: file integrity, annotation-structure validation, magic-byte spoofing detection, duplicate detection, and a 14-dimension scorer with multi-oracle agreement. Datasets scoring below the publication threshold are auto-rejected with feedback for the seller.
Yes. Upload, pass verification, set your price, and earn 85% per sale. No listing fees. The automated pipeline runs a full LQS pass and returns actionable feedback if your dataset needs work before publication.
Each dataset specifies its own license — Commercial, Research-only, CC BY, CC BY-NC, or MIT. Filter by license on the browse page. Commercial licenses allow use in production AI systems, including model weights distributed downstream.
The LQS cert is designed to drop into Article 10 data-governance documentation. The subgroup_equity dimension, per-dim 95% confidence intervals, and signed provenance chain address the Article 10 requirements on representativeness, relevance, and bias examination.

Browse all computer vision datasets.

Live marketplace with filters for LQS score, contamination-clean flag, format, and license. Or list your own vision dataset and start earning.

Browse datasets → Sell your dataset
Related categories
Autonomous Vehicles Medical Imaging NLP / Text Audio & Speech Financial Data