The open-source data labeling tool market is experiencing robust growth, driven by the increasing demand for high-quality training data in machine learning and artificial intelligence applications. The market's expansion is fueled by several factors: the rising adoption of AI across various sectors (including IT, automotive, healthcare, and finance), the need for cost-effective data annotation solutions, and the inherent flexibility and customization offered by open-source tools. While cloud-based solutions currently dominate the market due to scalability and accessibility, on-premise deployments remain significant, particularly for organizations with stringent data security requirements. The market's growth is further propelled by advancements in automation and semi-supervised learning techniques within data labeling, leading to increased efficiency and reduced annotation costs. Geographic distribution shows a strong concentration in North America and Europe, reflecting the higher adoption of AI technologies in these regions; however, Asia-Pacific is emerging as a rapidly growing market due to increasing investment in AI and the availability of a large workforce for data annotation. Despite the promising outlook, certain challenges restrain market growth. The complexity of implementing and maintaining open-source tools, along with the need for specialized technical expertise, can pose barriers to entry for smaller organizations. Furthermore, the quality control and data governance aspects of open-source annotation require careful consideration. The potential for data bias and the need for robust validation processes necessitate a strategic approach to ensure data accuracy and reliability. Competition is intensifying with both established and emerging players vying for market share, forcing companies to focus on differentiation through innovation and specialized functionalities within their tools. The market is anticipated to maintain a healthy growth trajectory in the coming years, with increasing adoption across diverse sectors and geographical regions. The continued advancements in automation and the growing emphasis on data quality will be key drivers of future market expansion.

According to Cognitive Market Research, the global Ai Training Data market size is USD 1865.2 million in 2023 and will expand at a compound annual growth rate (CAGR) of 23.50% from 2023 to 2030.

The demand for Ai Training Data is rising due to the rising demand for labelled data and diversification of AI applications.
Demand for Image/Video remains higher in the Ai Training Data market.
The Healthcare category held the highest Ai Training Data market revenue share in 2023.
North American Ai Training Data will continue to lead, whereas the Asia-Pacific Ai Training Data market will experience the most substantial growth until 2030.

Market Dynamics of AI Training Data Market

Key Drivers of AI Training Data Market

Rising Demand for Industry-Specific Datasets to Provide Viable Market Output

A key driver in the AI Training Data market is the escalating demand for industry-specific datasets. As businesses across sectors increasingly adopt AI applications, the need for highly specialized and domain-specific training data becomes critical. Industries such as healthcare, finance, and automotive require datasets that reflect the nuances and complexities unique to their domains. This demand fuels the growth of providers offering curated datasets tailored to specific industries, ensuring that AI models are trained with relevant and representative data, leading to enhanced performance and accuracy in diverse applications.

In July 2021, Amazon and Hugging Face, a provider of open-source natural language processing (NLP) technologies, have collaborated. The objective of this partnership was to accelerate the deployment of sophisticated NLP capabilities while making it easier for businesses to use cutting-edge machine-learning models. Following this partnership, Hugging Face will suggest Amazon Web Services as a cloud service provider for its clients.

(Source: about:blank)

Advancements in Data Labelling Technologies to Propel Market Growth

The continuous advancements in data labelling technologies serve as another significant driver for the AI Training Data market. Efficient and accurate labelling is essential for training robust AI models. Innovations in automated and semi-automated labelling tools, leveraging techniques like computer vision and natural language processing, streamline the data annotation process. These technologies not only improve the speed and scalability of dataset preparation but also contribute to the overall quality and consistency of labelled data. The adoption of advanced labelling solutions addresses industry challenges related to data annotation, driving the market forward amidst the increasing demand for high-quality training data.

In June 2021, Scale AI and MIT Media Lab, a Massachusetts Institute of Technology research centre, began working together. To help doctors treat patients more effectively, this cooperation attempted to utilize ML in healthcare.

www.ncbi.nlm.nih.gov/pmc/articles/PMC7325854/

Restraint Factors Of AI Training Data Market

Data Privacy and Security Concerns to Restrict Market Growth

A significant restraint in the AI Training Data market is the growing concern over data privacy and security. As the demand for diverse and expansive datasets rises, so does the need for sensitive information. However, the collection and utilization of personal or proprietary data raise ethical and privacy issues. Companies and data providers face challenges in ensuring compliance with regulations and safeguarding against unauthorized access or misuse of sensitive information. Addressing these concerns becomes imperative to gain user trust and navigate the evolving landscape of data protection laws, which, in turn, poses a restraint on the smooth progression of the AI Training Data market.

How did COVID–19 impact the Ai Training Data market?

The COVID-19 pandemic has had a multifaceted impact on the AI Training Data market. While the demand for AI solutions has accelerated across industries, the availability and collection of training data faced challenges. The pandemic disrupted traditional data collection methods, leading to a slowdown in the generation of labeled datasets due to restrictions on physical operations. Simultaneously, the surge in remote work and the increased reliance on AI-driven technologies for various applications fueled the need for diverse and relevant training data. This duali...

The data collection and labeling market is experiencing robust growth, fueled by the escalating demand for high-quality training data in artificial intelligence (AI) and machine learning (ML) applications. The market, estimated at $15 billion in 2025, is projected to achieve a Compound Annual Growth Rate (CAGR) of 25% over the forecast period (2025-2033), reaching approximately $75 billion by 2033. This expansion is primarily driven by the increasing adoption of AI across diverse sectors, including healthcare (medical image analysis, drug discovery), automotive (autonomous driving systems), finance (fraud detection, risk assessment), and retail (personalized recommendations, inventory management). The rising complexity of AI models and the need for more diverse and nuanced datasets are significant contributing factors to this growth. Furthermore, advancements in data annotation tools and techniques, such as active learning and synthetic data generation, are streamlining the data labeling process and making it more cost-effective. However, challenges remain. Data privacy concerns and regulations like GDPR necessitate robust data security measures, adding to the cost and complexity of data collection and labeling. The shortage of skilled data annotators also hinders market growth, necessitating investments in training and upskilling programs. Despite these restraints, the market’s inherent potential, coupled with ongoing technological advancements and increased industry investments, ensures sustained expansion in the coming years. Geographic distribution shows strong concentration in North America and Europe initially, but Asia-Pacific is poised for rapid growth due to increasing AI adoption and the availability of a large workforce. This makes strategic partnerships and global expansion crucial for market players aiming for long-term success.

1. Overview This dataset is a collection of high view traffic images in multiple scenes, backgrounds and lighting conditions that are ready to use for optimizing the accuracy of computer vision models. All of the contents is sourced from PIXTA's stock library of 100M+ Asian-featured images and videos. PIXTA is the largest platform of visual materials in the Asia Pacific region offering fully-managed services, high quality contents and data, and powerful tools for businesses & organisations to enable their creative and machine learning projects.

2. Use case This dataset is used for AI solutions training & testing in various cases: Traffic monitoring, Traffic camera system, Vehicle flow estimation,... Each data set is supported by both AI and human review process to ensure labelling consistency and accuracy. Contact us for more custom datasets.

3. About PIXTA PIXTASTOCK is the largest Asian-featured stock platform providing data, contents, tools and services since 2005. PIXTA experiences 15 years of integrating advanced AI technology in managing, curating, processing over 100M visual materials and serving global leading brands for their creative and data demands. Visit us at https://www.pixta.ai/ for more details.

Leaves from genetically unique Juglans regia plants were scanned using X-ray micro-computed tomography (microCT) on the X-ray μCT beamline (8.3.2) at the Advanced Light Source (ALS) in Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA USA). Soil samples were collected in Fall of 2017 from the riparian oak forest located at the Russell Ranch Sustainable Agricultural Institute at the University of California Davis. The soil was sieved through a 2 mm mesh and was air dried before imaging. A single soil aggregate was scanned at 23 keV using the 10x objective lens with a pixel resolution of 650 nanometers on beamline 8.3.2 at the ALS. Additionally, a drought stressed almond flower bud (Prunus dulcis) from a plant housed at the University of California, Davis, was scanned using a 4x lens with a pixel resolution of 1.72 µm on beamline 8.3.2 at the ALS Raw tomographic image data was reconstructed using TomoPy. Reconstructions were converted to 8-bit tif or png format using ImageJ or the PIL package in Python before further processing. Images were annotated using Intel’s Computer Vision Annotation Tool (CVAT) and ImageJ. Both CVAT and ImageJ are free to use and open source. Leaf images were annotated in following Théroux-Rancourt et al. (2020). Specifically, Hand labeling was done directly in ImageJ by drawing around each tissue; with 5 images annotated per leaf. Care was taken to cover a range of anatomical variation to help improve the generalizability of the models to other leaves. All slices were labeled by Dr. Mina Momayyezi and Fiona Duong.To annotate the flower bud and soil aggregate, images were imported into CVAT. The exterior border of the bud (i.e. bud scales) and flower were annotated in CVAT and exported as masks. Similarly, the exterior of the soil aggregate and particulate organic matter identified by eye were annotated in CVAT and exported as masks. To annotate air spaces in both the bud and soil aggregate, images were imported into ImageJ. A gaussian blur was applied to the image to decrease noise and then the air space was segmented using thresholding. After applying the threshold, the selected air space region was converted to a binary image with white representing the air space and black representing everything else. This binary image was overlaid upon the original image and the air space within the flower bud and aggregate was selected using the “free hand” tool. Air space outside of the region of interest for both image sets was eliminated. The quality of the air space annotation was then visually inspected for accuracy against the underlying original image; incomplete annotations were corrected using the brush or pencil tool to paint missing air space white and incorrectly identified air space black. Once the annotation was satisfactorily corrected, the binary image of the air space was saved. Finally, the annotations of the bud and flower or aggregate and organic matter were opened in ImageJ and the associated air space mask was overlaid on top of them forming a three-layer mask suitable for training the fully convolutional network. All labeling of the soil aggregate and soil aggregate images was done by Dr. Devin Rippner. These images and annotations are for training deep learning models to identify different constituents in leaves, almond buds, and soil aggregates Limitations: For the walnut leaves, some tissues (stomata, etc.) are not labeled and only represent a small portion of a full leaf. Similarly, both the almond bud and the aggregate represent just one single sample of each. The bud tissues are only divided up into buds scales, flower, and air space. Many other tissues remain unlabeled. For the soil aggregate annotated labels are done by eye with no actual chemical information. Therefore particulate organic matter identification may be incorrect. Resources in this dataset:Resource Title: Annotated X-ray CT images and masks of a Forest Soil Aggregate. File Name: forest_soil_images_masks_for_testing_training.zipResource Description: This aggregate was collected from the riparian oak forest at the Russell Ranch Sustainable Agricultural Facility. The aggreagate was scanned using X-ray micro-computed tomography (microCT) on the X-ray μCT beamline (8.3.2) at the Advanced Light Source (ALS) in Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA USA) using the 10x objective lens with a pixel resolution of 650 nanometers. For masks, the background has a value of 0,0,0; pores spaces have a value of 250,250, 250; mineral solids have a value= 128,0,0; and particulate organic matter has a value of = 000,128,000. These files were used for training a model to segment the forest soil aggregate and for testing the accuracy, precision, recall, and f1 score of the model.Resource Title: Annotated X-ray CT images and masks of an Almond bud (P. Dulcis). File Name: Almond_bud_tube_D_P6_training_testing_images_and_masks.zipResource Description: Drought stressed almond flower bud (Prunis dulcis) from a plant housed at the University of California, Davis, was scanned by X-ray micro-computed tomography (microCT) on the X-ray μCT beamline (8.3.2) at the Advanced Light Source (ALS) in Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA USA) using the 4x lens with a pixel resolution of 1.72 µm using. For masks, the background has a value of 0,0,0; air spaces have a value of 255,255, 255; bud scales have a value= 128,0,0; and flower tissues have a value of = 000,128,000. These files were used for training a model to segment the almond bud and for testing the accuracy, precision, recall, and f1 score of the model.Resource Software Recommended: Fiji (ImageJ),url: https://imagej.net/software/fiji/downloads Resource Title: Annotated X-ray CT images and masks of Walnut leaves (J. Regia) . File Name: 6_leaf_training_testing_images_and_masks_for_paper.zipResource Description: Stems were collected from genetically unique J. regia accessions at the 117 USDA-ARS-NCGR in Wolfskill Experimental Orchard, Winters, California USA to use as scion, and were grafted by Sierra Gold Nursery onto a commonly used commercial rootstock, RX1 (J. microcarpa × J. regia). We used a common rootstock to eliminate any own-root effects and to simulate conditions for a commercial walnut orchard setting, where rootstocks are commonly used. The grafted saplings were repotted and transferred to the Armstrong lathe house facility at the University of California, Davis in June 2019, and kept under natural light and temperature. Leaves from each accession and treatment were scanned using X-ray micro-computed tomography (microCT) on the X-ray μCT beamline (8.3.2) at the Advanced Light Source (ALS) in Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA USA) using the 10x objective lens with a pixel resolution of 650 nanometers. For masks, the background has a value of 170,170,170; Epidermis value= 85,85,85; Mesophyll value= 0,0,0; Bundle Sheath Extension value= 152,152,152; Vein value= 220,220,220; Air value = 255,255,255.Resource Software Recommended: Fiji (ImageJ),url: https://imagej.net/software/fiji/downloads

In 2020, more than 80 percent of the revenue of data annotation market were generated from English speaking regions, including the United States and the United Kingdom. The estimated data annotation market size serviced by India was around 250 million U.S. dollars in the same year. Data annotation is the process of labeling data in text, video, image, and other digital file formats.

1. Overview This dataset is a collection of 6,000+ images of mixed race human face with various expressions & emotions that are ready to use for optimizing the accuracy of computer vision models. All of the contents is sourced from PIXTA's stock library of 100M+ Asian-featured images and videos. PIXTA is the largest platform of visual materials in the Asia Pacific region offering fully-managed services, high quality contents and data, and powerful tools for businesses & organisations to enable their creative and machine learning projects.

2. The data set This dataset contains 6,000+ images of face emotion. Each data set is supported by both AI and human review process to ensure labelling consistency and accuracy. Contact us for more custom datasets.

3. About PIXTA PIXTASTOCK is the largest Asian-featured stock platform providing data, contents, tools and services since 2005. PIXTA experiences 15 years of integrating advanced AI technology in managing, curating, processing over 100M visual materials and serving global leading brands for their creative and data demands. Visit us at https://www.pixta.ai/ or contact via our email contact@pixta.ai."

This small data set include more than 50 commercial data integration products and services. We gather three attributes manually from the official website: the price availability, the use of semantics, and the use of machine learning for automatic identification of the same entities in different datasets. These are done manually based on the information that was available in the official websites. This might not be accurate as it is done manually. Nonetheless, we collected these information to show the general trend of the gathered attributes.

Description

CloudSEN12 is a large dataset for cloud semantic understanding that consists of 9880 regions of interest (ROIs). Each ROI has five 5090x5090 meters image patches (IPs) collected on different dates; we manually choose the images to guarantee that each IP inside an ROI matches one of the following cloud cover groups:

- clear (0%)

- low-cloudy (1% - 25%)

- almost clear (25% - 45%)

- mid-cloudy (45% - 65%)

- cloudy (65% >)

An IP is the core unit in CloudSEN12. Each IP contains data from Sentinel-2 optical levels 1C and 2A, Sentinel-1 Synthetic Aperture Radar (SAR), digital elevation model, surface water occurrence, land cover classes, and cloud mask results from eight cutting-edge cloud detection algorithms. Besides, in order to support standard, weakly, and self-/semi-supervised learning procedures, cloudSEN12 includes three distinct forms of hand-crafted labelling data: high-quality, scribble, and no annotation. Consequently, each ROI is randomly assigned to a different annotation group:

• 2000 ROIs with pixel-level annotation, where the average annotation time is 150 minutes (high-quality group).

• 2000 ROIs with scribble level annotation, where the annotation time is 15 minutes (scribble group).

• 5880 ROIs with annotation only in the cloud-free (0\%) image (no annotation group).

For high-quality labels, we use the Intelligence foR Image Segmentation\cite{iris2019} (IRIS) active learning technology, a system that combines human photo-interpretation and machine learning. For scribble, ground truth pixels were drawn using IRIS but without ML support. Finally, the no annotation dataset is generated automatically, with manual annotation only in the clear image patch. The dataset is already available here: https://shorturl.at/cgjtz. Check out our website https://cloudsen12.github.io/ for examples of how to download the dataset via STAC.

1. Overview This dataset is a collection of 5,000+ images of vehicle number plate position that are ready to use for optimizing the accuracy of computer vision models. All of the contents is sourced from PIXTA's stock library of 100M+ Asian-featured images and videos. PIXTA is the largest platform of visual materials in the Asia Pacific region offering fully-managed services, high quality contents and data, and powerful tools for businesses & organisations to enable their creative and machine learning projects.

2. Use case The 5,000+ images of vehicle number plate position could be used for various AI & Computer Vision models: Number Plate Recognition, Parking System, Surveillance Camera,... Each data set is supported by both AI and human review process to ensure labelling consistency and accuracy. Contact us for more custom datasets.

3. Annotation Annotation is available for this dataset on demand, including:

4. Bounding box

5. Classification

6. Segmentation ...

7. About PIXTA PIXTASTOCK is the largest Asian-featured stock platform providing data, contents, tools and services since 2005. PIXTA experiences 15 years of integrating advanced AI technology in managing, curating, processing over 100M visual materials and serving global leading brands for their creative and data demands. Visit us at https://www.pixta.ai/ or contact via our email contact@pixta.ai.

As single-cell chromatin accessibility profiling methods advance, scATAC-seq has become ever more important in the study of candidate regulatory genomic regions and their roles underlying developmental, evolutionary, and disease processes. At the same time, cell type annotation is critical in understanding the cellular composition of complex tissues and identifying potential novel cell types. However, most existing methods that can perform automated cell type annotation are designed to transfer labels from an annotated scRNA-seq data set to another scRNA-seq data set, and it is not clear whether these methods are adaptable to annotate scATAC-seq data. Several methods have been recently proposed for label transfer from scRNA-seq data to scATAC-seq data, but there is a lack of benchmarking study on the performance of these methods. Here, we evaluated the performance of five scATAC-seq annotation methods on both their classification accuracy and scalability using publicly available single-cell datasets from mouse and human tissues including brain, lung, kidney, PBMC, and BMMC. Using the BMMC data as basis, we further investigated the performance of these methods across different data sizes, mislabeling rates, sequencing depths and the number of cell types unique to scATAC-seq. Bridge integration, which is the only method that requires additional multimodal data and does not need gene activity calculation, was overall the best method and robust to changes in data size, mislabeling rate and sequencing depth. Conos was the most time and memory efficient method but performed the worst in terms of prediction accuracy. scJoint tended to assign cells to similar cell types and performed relatively poorly for complex datasets with deep annotations but performed better for datasets only with major label annotations. The performance of scGCN and Seurat v3 was moderate, but scGCN was the most time-consuming method and had the most similar performance to random classifiers for cell types unique to scATAC-seq.

The X-GENRE dataset comprises almost 3,000 web texts in English and Slovenian, manually-annotated with genre labels. The dataset allows for automated genre identification and genre analyses as well as other web corpora research. Inter alia, it was used for the development of the multilingual X-GENRE classifier (http://hdl.handle.net/11356/1961).

The X-GENRE dataset was constructed by merging three manually-annotated datasets by mapping the original schemata to the joint genre schema (the "X-GENRE schema"): 1) the Slovenian GINCO dataset (http://hdl.handle.net/11356/1467), 2) the English CORE dataset (https://github.com/TurkuNLP/CORE-corpus), and 3) the English FTD dataset (https://github.com/ssharoff/genre-keras). All of the original genre datasets are based on web corpora. The X-GENRE schema consists of 9 genre labels: Information/Explanation, News, Instruction, Opinion/Argumentation, Forum, Prose/Lyrical, Legal, Promotion and Other (refer to the README provided with the files for the details on the labels).

The dataset is separated into train, development and test split. The train split consists of 1,772 texts and 1,940,317 words, the development split of 592 texts and 798,025 words, and the test split of 592 texts and 583,595 words. The splits are stratified by labels. As the dataset consists of two English datasets and one Slovenian dataset, the distribution of texts in the two languages is roughly two to one: 2,063 English texts and 893 Slovenian texts.

The dataset is in JSONL format. It has the following attributes: text (text instance), labels (genre label), dataset (original manually-annotated genre dataset from which the instance was obtained – CORE, GINCO or FTD), and language (language of the text – Slovenian or English).

This work received funding from the European Union’s Connecting Europe Facility 2014–2020 – CEF Telecom – under Grant Agreement No. INEA/CEF/ICT/A2020/2278341. This communication reflects only the authors’ views. The Agency is not responsible for any use that may be made of the information it contains.

FSDKaggle2018 is an audio dataset containing 11,073 audio files annotated with 41 labels of the AudioSet Ontology. FSDKaggle2018 has been used for the DCASE Challenge 2018 Task 2, which was run as a Kaggle competition titled Freesound General-Purpose Audio Tagging Challenge.

Citation

If you use the FSDKaggle2018 dataset or part of it, please cite our DCASE 2018 paper:

Eduardo Fonseca, Manoj Plakal, Frederic Font, Daniel P. W. Ellis, Xavier Favory, Jordi Pons, Xavier Serra. "General-purpose Tagging of Freesound Audio with AudioSet Labels: Task Description, Dataset, and Baseline". Proceedings of the DCASE 2018 Workshop (2018)

You can also consider citing our ISMIR 2017 paper, which describes how we gathered the manual annotations included in FSDKaggle2018.

Eduardo Fonseca, Jordi Pons, Xavier Favory, Frederic Font, Dmitry Bogdanov, Andres Ferraro, Sergio Oramas, Alastair Porter, and Xavier Serra, "Freesound Datasets: A Platform for the Creation of Open Audio Datasets", In Proceedings of the 18th International Society for Music Information Retrieval Conference, Suzhou, China, 2017

Contact

You are welcome to contact Eduardo Fonseca should you have any questions at eduardo.fonseca@upf.edu.

About this dataset

Freesound Dataset Kaggle 2018 (or FSDKaggle2018 for short) is an audio dataset containing 11,073 audio files annotated with 41 labels of the AudioSet Ontology [1]. FSDKaggle2018 has been used for the Task 2 of the Detection and Classification of Acoustic Scenes and Events (DCASE) Challenge 2018. Please visit the DCASE2018 Challenge Task 2 website for more information. This Task was hosted on the Kaggle platform as a competition titled Freesound General-Purpose Audio Tagging Challenge. It was organized by researchers from the Music Technology Group of Universitat Pompeu Fabra, and from Google Research’s Machine Perception Team.

The goal of this competition was to build an audio tagging system that can categorize an audio clip as belonging to one of a set of 41 diverse categories drawn from the AudioSet Ontology.

All audio samples in this dataset are gathered from Freesound [2] and are provided here as uncompressed PCM 16 bit, 44.1 kHz, mono audio files. Note that because Freesound content is collaboratively contributed, recording quality and techniques can vary widely.

The ground truth data provided in this dataset has been obtained after a data labeling process which is described below in the Data labeling process section. FSDKaggle2018 clips are unequally distributed in the following 41 categories of the AudioSet Ontology:

"Acoustic_guitar", "Applause", "Bark", "Bass_drum", "Burping_or_eructation", "Bus", "Cello", "Chime", "Clarinet", "Computer_keyboard", "Cough", "Cowbell", "Double_bass", "Drawer_open_or_close", "Electric_piano", "Fart", "Finger_snapping", "Fireworks", "Flute", "Glockenspiel", "Gong", "Gunshot_or_gunfire", "Harmonica", "Hi-hat", "Keys_jangling", "Knock", "Laughter", "Meow", "Microwave_oven", "Oboe", "Saxophone", "Scissors", "Shatter", "Snare_drum", "Squeak", "Tambourine", "Tearing", "Telephone", "Trumpet", "Violin_or_fiddle", "Writing".

Some other relevant characteristics of FSDKaggle2018:

• The dataset is split into a train set and a test set.

• The train set is meant to be for system development and includes ~9.5k samples unequally distributed among 41 categories. The minimum number of audio samples per category in the train set is 94, and the maximum 300. The duration of the audio samples ranges from 300ms to 30s due to the diversity of the sound categories and the preferences of Freesound users when recording sounds. The total duration of the train set is roughly 18h.

• Out of the ~9.5k samples from the train set, ~3.7k have manually-verified ground truth annotations and ~5.8k have non-verified annotations. The non-verified annotations of the train set have a quality estimate of at least 65-70% in each category. Checkout the Data labeling process section below for more information about this aspect.

• Non-verified annotations in the train set are properly flagged in train.csv so that participants can opt to use this information during the development of their systems.

• The test set is composed of 1.6k samples with manually-verified annotations and with a similar category distribution than that of the train set. The total duration of the test set is roughly 2h.

• All audio samples in this dataset have a single label (i.e. are only annotated with one label). Checkout the Data labeling process section below for more information about this aspect. A single label should be predicted for each file in the test set.

Data labeling process

The data labeling process started from a manual mapping between Freesound tags and AudioSet Ontology categories (or labels), which was carried out by researchers at the Music Technology Group, Universitat Pompeu Fabra, Barcelona. Using this mapping, a number of Freesound audio samples were automatically annotated with labels from the AudioSet Ontology. These annotations can be understood as weak labels since they express the presence of a sound category in an audio sample.

Then, a data validation process was carried out in which a number of participants did listen to the annotated sounds and manually assessed the presence/absence of an automatically assigned sound category, according to the AudioSet category description.

Audio samples in FSDKaggle2018 are only annotated with a single ground truth label (see train.csv). A total of 3,710 annotations included in the train set of FSDKaggle2018 are annotations that have been manually validated as present and predominant (some with inter-annotator agreement but not all of them). This means that in most cases there is no additional acoustic material other than the labeled category. In few cases there may be some additional sound events, but these additional events won't belong to any of the 41 categories of FSDKaggle2018.

The rest of the annotations have not been manually validated and therefore some of them could be inaccurate. Nonetheless, we have estimated that at least 65-70% of the non-verified annotations per category in the train set are indeed correct. It can happen that some of these non-verified audio samples present several sound sources even though only one label is provided as ground truth. These additional sources are typically out of the set of the 41 categories, but in a few cases they could be within.

More details about the data labeling process can be found in [3].

License

FSDKaggle2018 has licenses at two different levels, as explained next.

All sounds in Freesound are released under Creative Commons (CC) licenses, and each audio clip has its own license as defined by the audio clip uploader in Freesound. For attribution purposes and to facilitate attribution of these files to third parties, we include a relation of the audio clips included in FSDKaggle2018 and their corresponding license. The licenses are specified in the files train_post_competition.csv and test_post_competition_scoring_clips.csv.

In addition, FSDKaggle2018 as a whole is the result of a curation process and it has an additional license. FSDKaggle2018 is released under CC-BY. This license is specified in the LICENSE-DATASET file downloaded with the FSDKaggle2018.doc zip file.

Files

FSDKaggle2018 can be downloaded as a series of zip files with the following directory structure:

root
│

└───FSDKaggle2018.audio_train/            Audio clips in the train set
│

└───FSDKaggle2018.audio_test/            Audio clips in the test set
│

└───FSDKaggle2018.meta/               Files for evaluation setup
│  │

│  └───train_post_competition.csv          Data split and ground truth for the train set
│  │

│  └───test_post_competition_scoring_clips.csv   Ground truth for the test set

│

└───FSDKaggle2018.doc/
  │

  └───README.md                  The dataset description file you are reading
  │

  └───LICENSE-DATASET             

Institute For the Study of Contemporary Antisemitism (ISCA) at Indiana University Dataset:

The ISCA project has compiled this dataset using an annotation portal, which was used to label tweets as either antisemitic or non-antisemitic, among other labels. Please note that the annotation was done with live data, including images and the context, such as threads. The original data was sourced from annotationportal.com.

Content:

This dataset contains 6,941 tweets that cover a wide range of topics common in conversations about Jews, Israel, and antisemitism between January 2019 and December 2021. The dataset is drawn from representative samples during this period with relevant keywords. 1,250 tweets (18%) meet the IHRA definition of antisemitic messages.

The dataset has been compiled within the ISCA project using an annotation portal to label tweets as either antisemitic or non-antisemitic. The original data was sourced from annotationportal.com.

The tweets' distribution of all messages by year is as follows: 1,499 (22%) from 2019, 3,716 (54%) from 2020, and 1,726 (25%) from 2021. 4,605 (66%) contain the keyword "Jews," 1,524 (22%) include "Israel," 529 (8%) feature the derogatory term "ZioNazi*," and 283 (4%) use the slur "K---s." Some tweets may contain multiple keywords.

483 out of the 4,605 tweets with the keyword "Jews" (11%) and 203 out of the 1,524 tweets with the keyword "Israel" (13%) were classified as antisemitic. 97 out of the 283 tweets using the antisemitic slur "K---s" (34%) are antisemitic. Interestingly, many tweets featuring the slur "K---s" actually call out its usage. In contrast, the majority of tweets with the derogatory term "ZioNazi*" are antisemitic, with 467 out of 529 (88%) being classified as such.

File Description:

The dataset is provided in a csv file format, with each row representing a single message, including replies, quotes, and retweets. The file contains the following columns:

‘TweetID’: Represents the tweet ID.

‘Username’: Represents the username who published the tweet.

‘Text’: Represents the full text of the tweet (not pre-processed).

‘CreateDate’: Represents the date the tweet was created.

‘Biased’: Represents the labeled by our annotations if the tweet is antisemitic or non-antisemitic.

‘Keyword’: Represents the keyword that was used in the query. The keyword can be in the text, including mentioned names, or the username.

Licences

Data is published under the terms of the "Creative Commons Attribution 4.0 International" licence (https://creativecommons.org/licenses/by/4.0)

R code is published under the terms of the "MIT" licence (https://opensource.org/licenses/MIT) ‘

Acknowledgements

We are grateful for the support of Indiana University’s Observatory on Social Media (OSoMe) (Davis et al. 2016) and the contributions and annotations of all team members in our Social Media & Hate Research Lab at Indiana University’s Institute for the Study of Contemporary Antisemitism, especially Grace Bland, Elisha S. Breton, Kathryn Cooper, Robin Forstenhäusler, Sophie von Máriássy, Mabel Poindexter, Jenna Solomon, Clara Schilling, and Victor Tschiskale.

This work used Jetstream2 at Indiana University through allocation HUM200003 from the Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support (ACCESS) program, which is supported by National Science Foundation grants #2138259, #2138286, #2138307, #2137603, and #2138296.

Collection of 10,000+ images of traffic scene from low view that are ready to use for optimizing the accuracy of computer vision models.

This dataset consists of sentences extracted from BGS memoirs, DECC/OGA onshore hydrocarbons well reports and Mineral Reconnaissance Programme (MRP) reports. The sentences have been annotated to enable the dataset to be used as labelled training data for a Named Entity Recognition model and Entity Relation Extraction model, both of which are Natural Language Processing (NLP) techniques that assist with extracting structured data from unstructured text. The entities of interest are rock formations, geological ages, rock types, physical properties and locations, with inter-relations such as overlies, observedIn. The entity labels for rock formations and geological ages in the BGS memoirs were an extract from earlier published work https://github.com/BritishGeologicalSurvey/geo-ner-model https://zenodo.org/records/4181488 . The data can be used to fine tune a pre-trained large language model using transfer learning, to create a model that can be used in inference mode to automatically create the labels, thereby creating structured data useful for geological modelling and subsurface characterisation. The data is provided in JSONL(Relation) format which is the export format from doccano open source text annotation software (https://doccano.github.io/doccano/) used to create the labels. The source documents are already publicly available, but the MRP and DECC reports are only published in pdf image form. These latter documents had to undergo OCR and resulted in lower quality text and a lower quality training data. The majority of the labelled data is from the higher quality BGS memoirs text. The dataset is a proof of concept. Minimal peer review of the labelling has been conducted so this should not be treated as a gold standard labelled dataset, and it is of insufficient volume to build a performant model. The development of this training data and the text processing scripts were supported by a grant from UK Government Office for Technology Transfer (GOTT) Knowledge Asset Grant Fund Project 10083604

Data DescriptionThe DIPSER dataset is designed to assess student attention and emotion in in-person classroom settings, consisting of RGB camera data, smartwatch sensor data, and labeled attention and emotion metrics. It includes multiple camera angles per student to capture posture and facial expressions, complemented by smartwatch data for inertial and biometric metrics. Attention and emotion labels are derived from self-reports and expert evaluations. The dataset includes diverse demographic groups, with data collected in real-world classroom environments, facilitating the training of machine learning models for predicting attention and correlating it with emotional states.Data Collection and Generation ProceduresThe dataset was collected in a natural classroom environment at the University of Alicante, Spain. The recording setup consisted of six general cameras positioned to capture the overall classroom context and individual cameras placed at each student’s desk. Additionally, smartwatches were used to collect biometric data, such as heart rate, accelerometer, and gyroscope readings.Experimental SessionsNine distinct educational activities were designed to ensure a comprehensive range of engagement scenarios:News Reading – Students read projected or device-displayed news.Brainstorming Session – Idea generation for problem-solving.Lecture – Passive listening to an instructor-led session.Information Organization – Synthesizing information from different sources.Lecture Test – Assessment of lecture content via mobile devices.Individual Presentations – Students present their projects.Knowledge Test – Conducted using Kahoot.Robotics Experimentation – Hands-on session with robotics.MTINY Activity Design – Development of educational activities with computational thinking.Technical SpecificationsRGB Cameras: Individual cameras recorded at 640×480 pixels, while context cameras captured at 1280×720 pixels.Frame Rate: 9-10 FPS depending on the setup.Smartwatch Sensors: Collected heart rate, accelerometer, gyroscope, rotation vector, and light sensor data at a frequency of 1–100 Hz.Data Organization and FormatsThe dataset follows a structured directory format:/groupX/experimentY/subjectZ.zip Each subject-specific folder contains:images/ (individual facial images)watch_sensors/ (sensor readings in JSON format)labels/ (engagement & emotion annotations)metadata/ (subject demographics & session details)Annotations and LabelingEach data entry includes engagement levels (1-5) and emotional states (9 categories) based on both self-reported labels and evaluations by four independent experts. A custom annotation tool was developed to ensure consistency across evaluations.Missing Data and Data QualitySynchronization: A centralized server ensured time alignment across devices. Brightness changes were used to verify synchronization.Completeness: No major missing data, except for occasional random frame drops due to embedded device performance.Data Consistency: Uniform collection methodology across sessions, ensuring high reliability.Data Processing MethodsTo enhance usability, the dataset includes preprocessed bounding boxes for face, body, and hands, along with gaze estimation and head pose annotations. These were generated using YOLO, MediaPipe, and DeepFace.File Formats and AccessibilityImages: Stored in standard JPEG format.Sensor Data: Provided as structured JSON files.Labels: Available as CSV files with timestamps.The dataset is publicly available under the CC-BY license and can be accessed along with the necessary processing scripts via the DIPSER GitHub repository.Potential Errors and LimitationsDue to camera angles, some student movements may be out of frame in collaborative sessions.Lighting conditions vary slightly across experiments.Sensor latency variations are minimal but exist due to embedded device constraints.CitationIf you find this project helpful for your research, please cite our work using the following bibtex entry:@misc{marquezcarpintero2025dipserdatasetinpersonstudent1, title={DIPSER: A Dataset for In-Person Student1 Engagement Recognition in the Wild}, author={Luis Marquez-Carpintero and Sergio Suescun-Ferrandiz and Carolina Lorenzo Álvarez and Jorge Fernandez-Herrero and Diego Viejo and Rosabel Roig-Vila and Miguel Cazorla}, year={2025}, eprint={2502.20209}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2502.20209}, } Usage and ReproducibilityResearchers can utilize standard tools like OpenCV, TensorFlow, and PyTorch for analysis. The dataset supports research in machine learning, affective computing, and education analytics, offering a unique resource for engagement and attention studies in real-world classroom environments.

Overview

This dataset of medical misinformation was collected and is published by Kempelen Institute of Intelligent Technologies (KInIT). It consists of approx. 317k news articles and blog posts on medical topics published between January 1, 1998 and February 1, 2022 from a total of 207 reliable and unreliable sources. The dataset contains full-texts of the articles, their original source URL and other extracted metadata. If a source has a credibility score available (e.g., from Media Bias/Fact Check), it is also included in the form of annotation. Besides the articles, the dataset contains around 3.5k fact-checks and extracted verified medical claims with their unified veracity ratings published by fact-checking organisations such as Snopes or FullFact. Lastly and most importantly, the dataset contains 573 manually and more than 51k automatically labelled mappings between previously verified claims and the articles; mappings consist of two values: claim presence (i.e., whether a claim is contained in the given article) and article stance (i.e., whether the given article supports or rejects the claim or provides both sides of the argument).

The dataset is primarily intended to be used as a training and evaluation set for machine learning methods for claim presence detection and article stance classification, but it enables a range of other misinformation related tasks, such as misinformation characterisation or analyses of misinformation spreading.

Its novelty and our main contributions lie in (1) focus on medical news article and blog posts as opposed to social media posts or political discussions; (2) providing multiple modalities (beside full-texts of the articles, there are also images and videos), thus enabling research of multimodal approaches; (3) mapping of the articles to the fact-checked claims (with manual as well as predicted labels); (4) providing source credibility labels for 95% of all articles and other potential sources of weak labels that can be mined from the articles' content and metadata.

The dataset is associated with the research paper "Monant Medical Misinformation Dataset: Mapping Articles to Fact-Checked Claims" accepted and presented at ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR '22).

The accompanying Github repository provides a small static sample of the dataset and the dataset's descriptive analysis in a form of Jupyter notebooks.

Options to access the dataset

There are two ways how to get access to the dataset:

1. Static dump of the dataset available in the CSV format
2. Continuously updated dataset available via REST API

In order to obtain an access to the dataset (either to full static dump or REST API), please, request the access by following instructions provided below.

References

If you use this dataset in any publication, project, tool or in any other form, please, cite the following papers:

@inproceedings{SrbaMonantPlatform, author = {Srba, Ivan and Moro, Robert and Simko, Jakub and Sevcech, Jakub and Chuda, Daniela and Navrat, Pavol and Bielikova, Maria}, booktitle = {Proceedings of Workshop on Reducing Online Misinformation Exposure (ROME 2019)}, pages = {1--7}, title = {Monant: Universal and Extensible Platform for Monitoring, Detection and Mitigation of Antisocial Behavior}, year = {2019} }

@inproceedings{SrbaMonantMedicalDataset, author = {Srba, Ivan and Pecher, Branislav and Tomlein Matus and Moro, Robert and Stefancova, Elena and Simko, Jakub and Bielikova, Maria}, booktitle = {Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR '22)}, numpages = {11}, title = {Monant Medical Misinformation Dataset: Mapping Articles to Fact-Checked Claims}, year = {2022}, doi = {10.1145/3477495.3531726}, publisher = {Association for Computing Machinery}, address = {New York, NY, USA}, url = {https://doi.org/10.1145/3477495.3531726}, }

Dataset creation process

In order to create this dataset (and to continuously obtain new data), we used our research platform Monant. The Monant platform provides so called data providers to extract news articles/blogs from news/blog sites as well as fact-checking articles from fact-checking sites. General parsers (from RSS feeds, Wordpress sites, Google Fact Check Tool, etc.) as well as custom crawler and parsers were implemented (e.g., for fact checking site Snopes.com). All data is stored in the unified format in a central data storage.

Ethical considerations

The dataset was collected and is published for research purposes only. We collected only publicly available content of news/blog articles. The dataset contains identities of authors of the articles if they were stated in the original source; we left this information, since the presence of an author's name can be a strong credibility indicator. However, we anonymised the identities of the authors of discussion posts included in the dataset.

The main identified ethical issue related to the presented dataset lies in the risk of mislabelling of an article as supporting a false fact-checked claim and, to a lesser extent, in mislabelling an article as not containing a false claim or not supporting it when it actually does. To minimise these risks, we developed a labelling methodology and require an agreement of at least two independent annotators to assign a claim presence or article stance label to an article. It is also worth noting that we do not label an article as a whole as false or true. Nevertheless, we provide partial article-claim pair veracities based on the combination of claim presence and article stance labels.

As to the veracity labels of the fact-checked claims and the credibility (reliability) labels of the articles' sources, we take these from the fact-checking sites and external listings such as Media Bias/Fact Check as they are and refer to their methodologies for more details on how they were established.

Lastly, the dataset also contains automatically predicted labels of claim presence and article stance using our baselines described in the next section. These methods have their limitations and work with certain accuracy as reported in this paper. This should be taken into account when interpreting them.

Reporting mistakes in the dataset The mean to report considerable mistakes in raw collected data or in manual annotations is by creating a new issue in the accompanying Github repository. Alternately, general enquiries or requests can be sent at info [at] kinit.sk.

Dataset structure

Raw data

At first, the dataset contains so called raw data (i.e., data extracted by the Web monitoring module of Monant platform and stored in exactly the same form as they appear at the original websites). Raw data consist of articles from news sites and blogs (e.g. naturalnews.com), discussions attached to such articles, fact-checking articles from fact-checking portals (e.g. snopes.com). In addition, the dataset contains feedback (number of likes, shares, comments) provided by user on social network Facebook which is regularly extracted for all news/blogs articles.

Raw data are contained in these CSV files (and corresponding REST API endpoints):

sources.csv

articles.csv

article_media.csv

article_authors.csv

discussion_posts.csv

discussion_post_authors.csv

fact_checking_articles.csv

fact_checking_article_media.csv

claims.csv

feedback_facebook.csv

Note: Personal information about discussion posts' authors (name, website, gravatar) are anonymised.

Annotations

Secondly, the dataset contains so called annotations. Entity annotations describe the individual raw data entities (e.g., article, source). Relation annotations describe relation between two of such entities.

Each annotation is described by the following attributes:

category of annotation (annotation_category). Possible values: label (annotation corresponds to ground truth, determined by human experts) and prediction (annotation was created by means of AI method).

type of annotation (annotation_type_id). Example values: Source reliability (binary), Claim presence. The list of possible values can be obtained from enumeration in annotation_types.csv.

method which created annotation (method_id). Example values: Expert-based source reliability evaluation, Fact-checking article to claim transformation method. The list of possible values can be obtained from enumeration methods.csv.

its value (value). The value is stored in JSON format and its structure differs according to particular annotation type.

At the same time, annotations are associated with a particular object identified by:

entity type (parameter entity_type in case of entity annotations, or source_entity_type and target_entity_type in case of relation annotations). Possible values: sources, articles, fact-checking-articles.

entity id (parameter entity_id in case of entity annotations, or source_entity_id and target_entity_id in case of relation annotations).

The dataset provides specifically these entity annotations:

Source reliability (binary). Determines validity of source (website) at a binary scale with two options: reliable source and unreliable source.

Article veracity. Aggregated information about veracity from article-claim pairs.

The dataset provides specifically these relation annotations:

Fact-checking article to claim mapping. Determines mapping between fact-checking article and claim.

Claim presence. Determines presence of claim in article.

Claim stance. Determines stance of an article to a claim.

Annotations are contained in these CSV files (and corresponding REST API endpoints):

entity_annotations.csv

relation_annotations.csv

Note: Identification of human annotators authors (email provided in the annotation app) is anonymised.

This collection contains 100 soundscape recordings of 10 minutes duration, which have been annotated with 10,296 bounding box labels for 21 different bird species from the Western United States. The data were recorded in 2015 in the southern end of the Sierra Nevada mountain range in California, USA. This collection has been featured as test data in the 2020 BirdCLEF and Kaggle Birdcall Identification competition and can primarily be used for training and evaluation of machine learning algorithms.

Data collection

The recordings were made in Sequoia and Kings Canyon National Parks, two contiguous national parks in the southern Sierra Nevada mountain range in California, USA. The focus of the acoustic study was the high-elevation region of the Parks; specifically, the headwater lake basins above 3,000 km in elevation. The original intent of the study was to monitor seasonal activity of birds and bats at lakes containing trout and lakes without trout, because the cascading impacts of trout on the adjacent terrestrial zone remain poorly understood. Soundscapes were recorded for 24 h continuously at 10 lakes (5 fishless, 5 fish-containing) throughout Sequoia and Kings Canyon National Parks during June-September 2015. Song Meter SM2+ units (Wildlife Acoustics, USA) powered by custom-made solar panels were used to obviate the need to swap batteries, due to the recording locations being extremely difficult to access. Song Meters continuously recorded mono-channel, 16-bits uncompressed WAVE files at 48 kHz sampling rate. For this collection, recordings were resampled at 32 kHz and converted to FLAC.

Sampling and annotation protocol

A total of 100 10-minute segments of audio between July 9 and 12, 2015 from morning hours (06:10-09:10 PDT) from all 10 sites were selected at random. Annotators were asked to box every bird call they could recognize, ignoring those that are too faint or unidentifiable. Every sound that could not be confidently assigned an identity was reviewed with 1-2 other experts in bird identification. To minimize observer bias, all identifying information about the location, date and time of the recordings was hidden from the annotator. Raven Pro software was used to annotate the data. Provided labels contain full bird calls that are boxed in time and frequency. In this collection, we use eBird species codes as labels, following the 2021 eBird taxonomy (Clements list). Unidentifiable calls have been marked with “????” and were added as bounding box labels to the ground truth annotations. Parts of this dataset have previously been used in the 2020 BirdCLEF and Kaggle Birdcall Identification competition.

Files in this collection

Audio recordings can be accessed by downloading and extracting the “soundscape_data.zip” file. Soundscape recording filenames contain a sequential file ID, recording date and timestamp in PDT (UTC-7). As an example, the file “HSN_001_20150708_061805.flac” has sequential ID 001 and was recorded on July 8th 2015 at 06:18:05 PDT. Ground truth annotations are listed in “annotations.csv” where each line specifies the corresponding filename, start and end time in seconds, low and high frequency in Hertz and an eBird species code. These species codes can be assigned to scientific and common name of a species with the “species.csv” file. The approximate recording location with longitude and latitude can be found in the “recording_location.txt” file.

Acknowledgements

Compiling this extensive dataset was a major undertaking, and we are very thankful to the domain experts who helped to collect and manually annotate the data for this collection (individual contributors in alphabetic order): Anna Calderón, Thomas Hahn, Ruoshi Huang, Angelly Tovar

Bottle Labels

## Overview

Bottle Labels is a dataset for object detection tasks - it contains Labels annotations for 644 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [Public Domain license](https://creativecommons.org/licenses/Public Domain).