Loading...
Loading...
Image classification, object detection, segmentation, face recognition, OCR, image generation, video understanding
16,989 datasets
SPSLingAnc provides a mixed-methods analysis of complete fictional narratives in English, focusing on linguistic tokens that shape readers' conceptualization of storyworld possible selves. The dataset comprises 91,328 words from two novels, with 5,980 tagged tokens analyzed using MAXQDA 2022 software. It was authored by María-Ángeles Martínez and last updated on October 14, 2025.
17,091,657 face images of 360,232 distinct individuals comprise this collection for training face recognition models. The dataset was cleaned and released by yayoimizuha, with the latest update recorded in October 2025. It is designed to support state-of-the-art performance on benchmarks like IJB-C and MegaFace.
The Kinetics-700 dataset is a large-scale collection of YouTube video URLs for human action recognition, containing 700 distinct action classes. It is an extension of the original Kinetics dataset, with each video clip being approximately 10 seconds long and depicting a single human action. The dataset was uploaded by author 'bitmind' to the Hugging Face platform, with a last recorded update in October 2025.
High-resolution images of handwritten physics notes, including theoretical explanations, formulas, diagrams, derivations, and problem-solving steps. The dataset was created by HumynLabs and last updated on Hugging Face in October 2025. It is designed to support AI research in handwriting recognition and scientific document understanding.
PaddlePaddle published a demonstration dataset for visual-language optical character recognition on Hugging Face on 2025-11-27. The dataset likely contains examples for testing or demonstrating a PaddleOCR visual-language model. Its specific content, size, and structure are not detailed in the provided metadata.
270,000 PDF pages converted to plain-text using GPT-4.1 with a specialized prompting strategy to preserve natural reading order and born-digital content. Released by the Allen Institute for AI (allenai) in October 2025, this collection includes both the original PDF files and their corresponding high-fidelity OCR outputs. It serves as an improved successor to the olmOCR-mix-0225 dataset with cleaner processing and more consistent formatting.
1,000 document images with human-annotated quality scores and OCR text extracted by the Qwen2.5-VL-72B model. The dataset was created by Aslan-mingye and last updated on October 17, 2025. It is designed as a benchmark for evaluating OCR performance across diverse document types including academic papers, textbooks, and e-books in multiple languages.
Sixteen mudbrick samples from the site of Tell Zurghul/Nigin in Iraq, dating from the mid-5th to mid 3rd millennium BCE, were analyzed. The dataset contains raw data from empirical field tests assessing material composition, grain size, clay behavior, and cohesion. It was created by De Vito, Licia as part of the EnEAp project and last updated in October 2025.
36 items across 4 categories record Spanish R&D projects on media communication ethics from 2007 to 2022. The database maps university research productivity over 15 years, providing derived data from project analysis. It was authored by Polledo-Zulueta, Yenisley and last updated on October 14, -2025.
Dutch survey data from 1993 to 1996, collected by the Central Bureau of Statistics (CBS). It aims to provide a complete overview of victimization of common crimes, feelings of insecurity, precautionary measures, and the use of legal aid and police assistance. The target population is individuals aged 15 and older in private households in the Netherlands.
A YOLO-format dataset for object detection tasks. The dataset contains 16 classes representing pool balls numbered 0 through 15. It was created by author nhantu2107 and was last updated on Hugging Face on October 23, 2025.
176,999 programming conversations translated into Oromo from the Glaive Code Assistant v2 dataset. These multi-turn dialogues cover technical topics such as algorithms and data structures to support coding education for Oromo speakers.
1968 onward, this dataset contains Oregon workers' compensation claims counts and insurer performance metrics. It is provided by the Oregon Department of Consumer and Business Services and includes annual data on accepted and denied claims, fatality rates, and processing timeliness. The data supports the official state report on workers' compensation.
Mexico City crime victim data extracted from the open data portal of the Mexico City Government. The dataset was adapted for use in Tableau with OpenRefine and is used for a course on data visualization principles. The data was last updated on the platform on October 14,我们发现2025.
3,000 images and fine-grained alpha mattes serve as the foundation for zero-shot image matting benchmarks. These samples enable the development of models that require higher precision than standard segmentation masks for complex edges like hair or transparency.
ILSVRC 2012 (ImageNet 1K) contains over 1.2 million images categorized into 1,000 distinct object classes based on the WordNet hierarchy. Created by the ILSVRC team and released in 2012, it serves as the foundational benchmark for large-scale visual recognition and object classification.
French Cold War documents collected by Marc Trachtenberg for historical research. The collection covers the period between 1945 and 1959 and is organized into three folders for five-year intervals. It was harvested by QDR and last updated on October 20, 2025.
Marc Trachtenberg's collection of British Cold War documents supports his historical analysis, notably in 'A Constructed Peace'. The data covers the pivotal period from the end of WWII to 1964, organized into folders for five-year intervals and individual years. This particular project encompasses documents from the United Kingdom.
A cleaned version of the Aozora Bunko corpus processed with fugashi morphological analysis and the unidic-lite dictionary. Each morpheme was converted into hiragana, and all non-hiragana symbols were removed. The dataset was created by milano0017 and last updated on November 10, 2025.
MTADataset is a large-scale dataset designed for image inpainting. It contains images processed with Grounded-SAM to extract labels, bounding boxes, and masks, and uses LLaVA to generate detailed descriptions for approximately 5 masks per image. The dataset was created by huangjun12 and was last updated on October 23, 2025.