Loading...
Loading...
Image classification, object detection, segmentation, face recognition, OCR, image generation, video understanding
16,950 datasets
203,048 question-answer pairs accompany 2,507 video trajectories generated from 1,261 procedurally created indoor scenes. The dataset was created by Ellis Brown, Arijit Ray, Ranjay Krishna, Ross Girshick, Rob Fergus, and Saining Xie for training models in spatial video understanding. It was last updated in November 2025.
164 frame images extracted from Sora AI-generated videos, annotated for watermark detection. The dataset, created by LLinked and last updated in October 2025, is formatted for YOLOv11 object detection models. It is partitioned into 124 training, 21 validation, and 19 test images.
164 frame images extracted from Sora AI-generated videos are annotated for watermark detection. The dataset, created by LLinked and updated in October 2025, is formatted for YOLOv11 object detection. It is split into 124 training, 21 validation, and 19 test images.
MedSegDB is a collection of publicly available medical image datasets aggregated for academic research, curated by medicalai. The dataset page was last updated on November 8, 2025. It serves as a foundation for an open-world medical image segmentation model.
3 terabytes of raw airborne radar data were collected during the 2022/23 and 2023/24 Antarctic field seasons for the Center for Oldest Ice Exploration (COLDEX). The dataset includes processed radargrams in HDF5/Matlab and netCDF formats, as well as reprojected images in PNG/JPG formats, organized for non-expert access. Duncan A. Young is the author, and the data is hosted via the Texas Data Repository and the University of Kansas CReSIS data portal.
A text corpus derived from UK content on Wikisource, processed through optical character recognition (OCR) and cleaned. The dataset was published on huggingface by author JustQuiteMadMax and was last updated on 2025-12-15 20:51:07. Its specific size, row count, and license details are currently unknown.
Books Ocr Split V1 is a dataset uploaded by author oddadmix to the Hugging Face platform. The dataset was last updated on 2025-12-27 02:11:54. Its specific content, scale, and structure require verification after download.
PMoC3D is a real-world dataset of Paired Motion-Corrupted 3D brain MRI data. It consists of unprocessed undersampled measurement data for 27 motion-corrupted volumes from 8 subjects, each paired with a motion-free reference scan. The dataset was created by mli-lab and last updated on Hugging Face in November 2025.
AlvaroCavalcante released this hand and face detection dataset on GitHub, with the most recent update occurring in January 2026. It provides visual data specifically for sign language recognition tasks using TensorFlow-based object detection frameworks.
Research data from e-cienciaDatos Harvested Dataverse investigates the influence of common impurities on heterogeneous acid catalysts for producing bio-jet fuel precursors. The dataset, authored by Marta Paniagua and last updated in November 2025, compares sulfonic acid-based materials and commercial zeolites. It details catalyst deactivation and regeneration processes, identifying furfural as the most detrimental impurity.
Gil Martín, Manuel's Multi-view Leap2 Hand Pose Dataset (ML2HP Dataset) is a resource for hand pose recognition, captured using two Leap Motion Controller 2 devices. It contains 714,000 instances of 17 different hand poses, recorded from 21 subjects, and includes real images with 247 associated hand properties like landmark coordinates and velocities. The dataset was last updated on October 14, III.
Vokturz created a dataset containing 1,118 screenshots from SourceForge software projects. The images are paired with metadata and text extracted via the qwen/qwen3-vl-235b-a22b-instruct model. The dataset was last updated on November 8, 2025, and is intended for fine-tuning smaller Qwen3-VL models.
Comprising data from 13,051 children under five years old enrolled in a two-phase study across six Ugandan hospitals to assess a risk-differentiated discharge intervention for suspected sepsis. It includes clinical, social, and demographic variables collected at admission and follow-up data on mortality, health-seeking, and readmission up to six months post-discharge. The study compares a baseline phase (n=6,955) with an intervention phase (n=6,096).
491 MB of data linking every Unique Property Reference Number (UPRN) in Great Britain to statutory administrative, electoral, health, and other geographies. The ONS UPRN Directory is produced by ONS Geography and updated every six weeks to complement the Ordnance Survey AddressBase product. This snapshot is from February 2023.
VidChapters-7M contains 817,000 videos with associated Automatic Speech Recognition data and chapter annotations. The dataset was created by lucas-ventura for the CVPR 2025 paper on efficient chaptering in hour-long videos. It includes captions extracted using various sampling strategies and chapter titles with timestamps.
Karine-Huang developed T2I-CompBench in 2023 to provide a standardized framework for evaluating compositionality in text-to-image models. The benchmark includes curated text prompts and evaluation metrics published in NeurIPS 2023 and TPAMI to assess how models handle complex textual instructions.
1460 coded political actors across 290 presidential administrations in 20 Latin American countries. Scott Mainwaring commissioned archival research reports between 2008 and 2013 to measure normative regime preferences and policy radicalism. The data collection covers the period from 1944 to 2010, with Argentina and El Salvador extending back to 1916 and 1927.
STIPLAR is a real-world scene text image dataset containing Korean, Arabic, and Japanese text image pairs. The data was collected from MLT-2019 and web sources and is designed for fine-tuning the STELLAR model on low-resource languages. It was created by yongchoooon and last updated on Hugging Face in November 2025.
The Yiddish Synthetic Pangoline Dataset is a collection of synthetic Yiddish document images generated using a custom Pangoline text-to-image synthesis tool. It contains high-quality synthetic Yiddish text rendered as images, along with corresponding ground truth text and ALTO-XML layout annotations. The dataset was created by author johnlockejrr and was last updated on 2025-11-02.
A qualitative study deposited in October 2025 examines how search systems impact systematic searching. Data were collected from interviews with twelve systematic searchers and analyzed using reflexive thematic analysis. The dataset includes deidentified interview transcripts, recruitment materials, and coded themes for two planned publications.