Medical — newer sets
Journal article corpus (illustrated cases and captioned imaging) with a documented sublicense path. Confirm scope, redistribution limits, and exclusions before purchase.
5 million structured checkup records (counted by case), roughly balanced by sex and stratified by age. This is not the 2,000-set custom subset that includes original reports.
Custom subset with original reports, counted by set. For training that needs full report text and historical comparison. Not the 5-million structured library.
De-identified EMR text covering chief complaint, history, and admission/discharge fields. Confirm slice, de-identification rules, and exclusions before purchase.
Textbooks, guidelines, reviews, and drug labels as a medical knowledge source.
About 5 million 12-lead ECG waveforms and reports (counted by case), with automated analysis and physician-review status. Samples are controlled; authorization is confirmed in the project.
Dermoscopy, clinical lesion photos, and pathology reports paired for classification or detection. Counts, de-identification, and license limits are in the detail page.
Chest radiographs paired with reports for detection and classification. Modality, QC, and license limits are on the detail page.
5 million head CT studies (plain and contrast) with reports, covering infarction, hemorrhage, and intracranial tumors.
Fundus color photos and structured reports using ICDR diabetic-retinopathy staging.
Whole-slide images (20x and above) with full reports, including IHC and TNM stage.
5 million endoscopy videos with keyframes and reports, including polyp location, size, and type.
5 million annotated upper-abdominal multi-organ cancer cases covering liver, biliary tract, pancreas, spleen, adrenal, kidney, and nodes.
5 million pathology cases spanning the diagnostic workflow in image and text.
5 million difficult-case MDT texts for rare and complex disease training.
1 million longitudinal chronic-disease records with 7 or more admissions.
5 million QuPath region annotations with HCC, ICC, MID, TLS, and MVI labels.
1 million liver-lesion image annotations for cancer versus hemangioma.
1 million breast-tumor image annotations for screening and diagnosis.
20 million inpatient documents (text/PDF) for record QC and DRG/DIP.
5 million+ medical entities and 20 million+ triples, mapped to ICD-10/11 and SNOMED CT.
100 million+ medical NLP annotations across 12 task families and 66 subsets, including extraction, dialogue, and chain-of-thought.
20 million+ imaging samples across CT, MRI, X-ray, ultrasound, endoscopy, and pathology, with classification, detection, and segmentation labels.
3 million+ structured records across 26 priority diseases, including diabetes, stroke, and cancer, with 5+ years of follow-up.
Medical — established sets
1 million+ in-depth checkup reports with 800+ structured fields.
328,000 EMR, report, order, and abstract items annotated for NER and relation extraction.
52,800 DICOM studies with physician annotation.
52,800 multi-sequence HCC MRI studies with physician annotation.
A general imaging set across CT, X-ray, ultrasound, and MRI.
1 million+ annotated surgical frames with pixel labels for 24 anatomical structures and instruments.
Large OB/GYN clinical coverage from prenatal emergencies through postpartum care.
52,800 DICOM studies with physician annotation.
1 million+ tongue and face images plus 50,000+ professional annotations.
1 million+ annotated ultrasound frames.
1 million+ DICOM slices across 7 organs, thin-slice CT at 512×512 and above.
AI Coding datasets NEW
Agent evaluation and task environments HOT
DeepResearch and expert annotation NEW
Embodied intelligence datasets HOT
Egocentric behavior, humanoid imitation, factory, retail, home, and office data, plus an on-site capture-hardware catalog.
E6 / E2 / E8, G7 / G2-L, wrist camera, wheeled teleop bases, teleop kits, and grippers. Open the English matrix or the embodied page. Specification sheets are published in Chinese.
Egocentric video of daily, kitchen, office, and retail scenes with head IMU, gaze, and hand keypoints.
Humanoid imitation episodes (stereo RGB-D, proprioception, force). Sync and license are confirmed on the detail page.
Evaluation set for generative world models: physical consistency, object tracks, viewpoint change, and long rollouts.
Single-arm, dual-arm, mobile, and dexterous-hand embodiments in one action and coordinate convention.
Simulated bimanual assembly, carry, fold, and collaboration with depth, instance masks, and contact labels.
Contact-rich episodes with tactile, force/torque, and macro/micro vision. Samples are controlled, not a public download.
Factory screw, insertion, cable, and inspection tasks with multi-view video, tool pose, torque, and SOP text.
Parcel grasp, palletize, and sort data with 6D pose, conveyor motion, and damage labels.
Sweep, wipe, tidy, and sort tasks from bare-hand and wearable cameras across homes and lighting.
Front-of-house and kitchen tasks covering food state, service steps, and food-safety actions.
PharmaceuticalsNEW
Drugs, traditional Chinese medicine formulas, and device/consumable records.
5 million+ Chinese drug records across 15+ fields, including ATC level 5.
2 million+ TCM herbs, formulas, and hospital preparations, including properties, actions, and origin.
1 million+ device registrations and consumables, including class, intended use, and parameters.
Medical insuranceNEW
Settlement lists, DIP groups, DRG codes, and payment standards.
10 million+ settlement records across catalog, DIP group, DRG code, and cost structure.
International corporaNEW
10 languages and 5 million+ medical texts for translation and cross-lingual graphs.
5 million+ multilingual medical texts in 10 languages and 8 types, with ICD-11 / SNOMED CT tags.
Industrial economy
2D/3D CAD drawings for drawing understanding, BIM conversion, and industrial models.
Sales, lease, service, and employment contracts for review and risk models.
Financial statements, industry research, and financing events.
Supplier links, purchase orders, logistics nodes, and inventory.
Public livelihood
Housing expropriation, vouchers, and resettlement compensation.
Acquisition extent, compensation, title, and approval flow.
City operations, grid events, and public-service records.
AI + science
Hard math, physics, chemistry, and computer-science problems with solutions.
Standardized items and explanations for tutoring and diagnosis models.
Past papers, mocks, and response data for assessment and scoring research.
Abstracts, citations, and keywords for retrieval and survey generation.
Policy, patents, standards, and industry signals for science intelligence.
Expert-level multidisciplinary questions for measuring model limits.
AI + security
Samples of network attacks, social engineering, and prompt injection for red-team evaluation and safety alignment. This catalog does not include exploit procedures.
Detection-boundary and false-positive cases used to find blind spots. Evaluation framing only.
Labeled overflow examples for security evaluation and static-analysis research. Not an exploit guide.
Path-traversal payloads for file-handling detection evaluation.
Multi-source logs, traffic, and alerts for SOC triage and ATT&CK mapping research.
Legal
AI + law notes (2025–2026)
Langhui legal data
Langhui supplies legal Q&A, judgments, contract review, retrieval, and reasoning data under a documented license.
Maritime-law multiple choice, case analysis, and precedent reasoning.
Complaints, answers, and appeals for document generation and format review.
Chinese judgments with cause, facts, reasoning, and outcome.
National and local statutes, administrative rules, and judicial interpretations.
Contracts with risk-clause labels for review and extraction.
Common legal questions paired with lawyer answers.
Summaries and issue spotting from judgments. Not a legal-advice product.
Parties, statutes, courts, and causes of action labeled in legal text.
Chains from facts to applicable law and outcome.
Chinese-English and other parallel legal text.
Typical uses
Culture and art
High-resolution heritage images with transcriptions for digitization and calligraphy models.
World models
Multi-view photos with camera pose, intrinsics, and sparse reconstruction for NeRF / 3DGS.
Orthophoto and oblique drone imagery with POS, RTK, and gimbal angle.
Vehicle, aerial, and terrestrial LiDAR with intensity, semantics, and trajectory.
Photos paired with textured 3D models for image-to-3D generation.
High-detail scenes with semantic and instance labels and interactive objects.
Part-level semantics, assembly order, and constraints for digital twins and assembly planning.
Need more industry datasets?Scope a custom data plan
This page lists 116 dataset cards across 15 domains. The medical specialty page lists 96 catalog items (specification packs). That is a different count from the 35 medical dataset cards here. For custom capture or annotation, send a brief and we will confirm scope and delivery.