Loading...
Loading...
Image classification, object detection, segmentation, face recognition, OCR, image generation, video understanding
16,937 datasets
María Sanz's dataset from e-cienciaDatos Harvested Dataverse, last updated in November 2025, presents experimental results for producing jet-fuel precursors. The data covers the aldol condensation of furfural with methyl-isobutyl ketone over defect-engineered UiO-66(Zr) catalysts. Catalyst characterization includes XRD, argon adsorption, TGA, TEM, and FTIR spectroscopy results.
Infinity-Doc-400K contains 400,000 real-world and synthetic scanned documents featuring diverse layout variations and structural annotations, released by infly in 2025. This dataset expands upon the previous 55K version to support document parsing model training across financial, medical, and academic domains. It is provided in Parquet format and includes both image and text modalities.
Metadata for the VGGSound dataset, annotated with video and audio prompts for the paper "Does Hearing Help Seeing? Investigating Audio-Video Joint Denoising for Video Generation". The dataset was created by author jianzongwu and the metadata was last updated on December 3, 2025. The repository contains annotated train and evaluation metadata, while the video files must be downloaded from a separate source.
A dataset for optical character recognition (OCR) tasks involving the Nepali language. It was published on the Hugging Face platform by the user gauravgiri and was last updated on January 6, 2026. The specific content, scale, and structure require verification after download.
UltraHR-100K is a large-scale, high-quality dataset introduced to address the absence of such data for ultra-high-resolution (UHR) text-to-image generation. It was created by author zhihefang and was last updated on the Hugging Face platform in November 2025. The dataset is intended to support progress in fine-grained detail synthesis for UHR scenarios.
169 Connecticut towns contributed property assessment data for a statewide 2024 collection effort. The GIS office compiled this information into a seamless parcel layer, cleaning inconsistencies from municipal submissions. This file was generated on September 28, 2024, from data collected throughout the year.
Satellitesegmentation contains paired Sentinel-2 RGB image tiles and corresponding integer-labeled land-cover masks derived from ESA WorldCover data. The dataset is prepared for semantic segmentation tasks, with source imagery from Sentinel-2 L2A via Microsoft Planetary Computer. It was authored by nikolkoo and last updated on 2025-11-26.
This collection of Jupyter Notebooks by Ultralytics provides implementation examples for computer vision tasks including object detection and instance segmentation. Updated in January 2026, the repository serves as the official documentation for YOLO11 and previous model iterations across medical and manufacturing domains.
5,082 parsed email threads extracted from U.S. House Oversight Committee documents and processed by notesbymuneeb in late 2024. It transforms raw OCR'd text into a structured format featuring cleaned message bodies, sender/recipient metadata, and timestamps.
Performance data for the Accountable Care Organization REACH Model, formerly the Global and Professional Direct Contracting Model, prior to settlement. The dataset includes information on risk arrangements, stop loss, capitation, savings rates, and quality results for participating ACOs. It is published by the U.S. Department of Health & Human Services.
U.S. Department of Health & Human Services data details usage, payments, and submitted charges for Medicare Durable Medical Equipment, Devices & Supplies. Information is organized by National Provider Identifier (NPI), Healthcare Common Procedure Coding System (HCPCS) code, and supplier rental indicator. The dataset was last updated in December 2025.
craneset maintains a collection of datasets related to Optical Character Recognition (OCR). The repository is licensed under the MIT license and was last updated on 2026-01-06. The specific contents, scale, and formats of the datasets are not detailed in the available metadata.
InnovisionLLC generated this high-quality dataset for the human injury biomechanics community. Version V1.0 was supported by NSF ACCESS and the US Department of Defense. The dataset was last updated on December 18, 2025.
A web-scale dataset introduced in the paper 'CommonForms: A Large, Diverse Dataset for Form Field Detection'. It contains roughly 55,000 document images annotated for form field detection, framing the task as object detection. The dataset was created by jbarrow and was last updated on Hugging Face in November 2025.
A large-scale multi-modal dataset for referring motion expression video segmentation, developed by FudanCVL. The dataset is associated with research papers published at ICCV 2023 and TPAMI 2025. The dataset page was last updated on 2025-11-27.
Multiple processed benchmark datasets across depth estimation and visual geometry categories provide standardized evaluation for the Depth Anything 3 model. The collection includes test sets for predicting spatially consistent geometry from arbitrary visual inputs across varying camera configurations.
A dataset created in partnership with Microsoft for a research project on positive representation in AI image generation software. Little People of America (LPA) was invited to help develop AI software designed to improve the respectful and accurate visual representation of people with dwarfism. The dataset was authored by LPA1957 and last updated on 2025-11-25.
2.6 million image-text pairs form this synthetic dataset for Persian Optical Character Recognition. Persian-OCR-Garshasp is designed for training and evaluating OCR and scene text recognition models for Persian text in the wild. The dataset was created by AliShafiee2003 and was last updated on Hugging Face in November 2025.
EgoPlan-Bench provides 741 samples for evaluating models on egocentric video understanding and action planning. Each sample contains a video file, a question about the next action, a task goal, and four multiple-choice options with a correct answer. The dataset is designed as a benchmark for tasks involving video and textual reasoning.
This dataset provides images for training Not Safe For Work detection models, focusing on artistic content. It is organized into five distinct classes representing different levels of appropriateness. The dataset was created by author deepghs and was last updated in November 2025.