Explore the booming data collection and labeling market, driven by AI advancements. Discover key growth drivers, market trends, and forecasts for 2025-2033, essential for AI development across IT, automotive, and healthcare.

Supervised machine learning methods for image analysis require large amounts of labelled training data to solve computer vision problems. The recent rise of deep learning algorithms for recognising image content has led to the emergence of many ad-hoc labelling tools. With this survey, we capture and systematise the commonalities as well as the distinctions between existing image labelling software. We perform a structured literature review to compile the underlying concepts and features of image labelling software such as annotation expressiveness and degree of automation. We structure the manual labelling task by its organisation of work, user interface design options, and user support techniques to derive a systematisation schema for this survey. Applying it to available software and the body of literature, enabled us to uncover several application archetypes and key domains such as image retrieval or instance identification in healthcare or television.

We present Code4ML: a Large-scale Dataset of annotated Machine Learning Code, a corpus of Python code snippets, competition summaries, and data descriptions from Kaggle.

The data is organized in a table structure. Code4ML includes several main objects: competitions information, raw code blocks collected form Kaggle and manually marked up snippets. Each table has a .csv format.

Each competition has the text description and metadata, reflecting competition and used dataset characteristics as well as evaluation metrics (competitions.csv). The corresponding datasets can be loaded using Kaggle API and data sources.

The code blocks themselves and their metadata are collected to the data frames concerning the publishing year of the initial kernels. The current version of the corpus includes two code blocks files: snippets from kernels up to the 2020 year (сode_blocks_upto_20.csv) and those from the 2021 year (сode_blocks_21.csv) with corresponding metadata. The corpus consists of 2 743 615 ML code blocks collected from 107 524 Jupyter notebooks.

Marked up code blocks have the following metadata: anonymized id, the format of the used data (for example, table or audio), the id of the semantic type, a flag for the code errors, the estimated relevance to the semantic class (from 1 to 5), the id of the parent notebook, and the name of the competition. The current version of the corpus has ~12 000 labeled snippets (markup_data_20220415.csv).

As marked up code blocks data contains the numeric id of the code block semantic type, we also provide a mapping from this number to semantic type and subclass (actual_graph_2022-06-01.csv).

The dataset can help solve various problems, including code synthesis from a prompt in natural language, code autocompletion, and semantic code classification.

Handwritten Digits Pixel Dataset - Documentation

Overview

The Handwritten Digits Pixel Dataset is a collection of numerical data representing handwritten digits from 0 to 9. Unlike image datasets that store actual image files, this dataset contains pixel intensity values arranged in a structured tabular format, making it ideal for machine learning and data analysis applications.

Dataset Description

Basic Information

• Format: CSV (Comma-Separated Values)
• Total Samples: [Number of rows based on your dataset]
• Features: 784 pixel columns (28×28 pixels) + 1 label column
• Label Range: Digits 0-9
• Pixel Value Range: 0-255 (grayscale intensity)

File Structure

Column Description

• label: The target variable representing the digit (0-9)
• pixel columns: 784 columns named in format [row]xcolumn
• Each pixel column contains integer values from 0-255 representing grayscale intensity

Data Characteristics

Label Distribution

The dataset contains handwritten digit samples with the following distribution:

• Digit 0: [X] samples
• Digit 1: [X] samples
• Digit 2: [X] samples
• Digit 3: [X] samples
• Digit 4: [X] samples
• Digit 5: [X] samples
• Digit 6: [X] samples
• Digit 7: [X] samples
• Digit 8: [X] samples
• Digit 9: [X] samples

(Note: Actual distribution counts would be calculated from your specific dataset)

Data Quality

• Missing Values: No missing values detected
• Data Type: All values are integers
• Normalization: Pixel values range from 0-255 (can be normalized to 0-1 for ML models)
• Consistency: Uniform 28×28 grid structure across all samples

Technical Specifications

Data Preprocessing Requirements

• Normalization: Scale pixel values from 0-255 to 0-1 range
• Reshaping: Convert 1D pixel arrays to 2D 28×28 matrices for visualization
• Train-Test Split: Recommended 80-20 or 70-30 split for model development

Recommended Machine Learning Approaches

Classification Algorithms:

• Random Forest
• Support Vector Machines (SVM)
• Neural Networks
• K-Nearest Neighbors (KNN)

Deep Learning Architectures:

• Convolutional Neural Networks (CNNs)
• Multi-layer Perceptrons (MLPs)

Dimensionality Reduction:

• PCA (Principal Component Analysis)
• t-SNE for visualization

Usage Examples

Loading the Dataset

import pandas as pd

# Load the dataset
df = pd.read_csv('/kaggle/input/handwritten_digits_pixel_dataset/mnist.csv')

# Separate features and labels
X = df.drop('label', axis=1)
y = df['label']

# Normalize pixel values
X_normalized = X / 255.0

Sharing cooking recipes is a great way to exchange culinary ideas and provide instructions for food preparation. However, categorizing raw recipes found online into appropriate food genres can be challenging due to a lack of adequate labeled data. In this study, we present a dataset named the “Assorted, Archetypal, and Annotated Two Million Extended (3A2M+) Cooking Recipe Dataset” that contains two million culinary recipes labeled in respective categories with extended named entities extracted from recipe descriptions. This collection of data includes various features such as title, NER, directions, and extended NER, as well as nine different labels representing genres including bakery, drinks, non-veg, vegetables, fast food, cereals, meals, sides, and fusions. The proposed pipeline named 3A2M+ extends the size of the Named Entity Recognition (NER) list to address missing named entities like heat, time or process from the recipe directions using two NER extraction tools. 3A2M+ dataset provides a comprehensive solution to the various challenging recipe-related tasks, including classification, named entity recognition, and recipe generation. Furthermore, we have demonstrated traditional machine learning, deep learning and pre-trained language models to classify the recipes into their corresponding genre and achieved an overall accuracy of 98.6%. Our investigation indicates that the title feature played a more significant role in classifying the genre.

These images and associated binary labels were collected from collaborators across multiple universities to serve as a diverse representation of biomedical images of vessel structures, for use in the training and validation of machine learning tools for vessel segmentation. The dataset contains images from a variety of imaging modalities, at different resolutions, using difference sources of contrast and featuring different organs/ pathologies. This data was use to train, test and validated a foundational model for 3D vessel segmentation, tUbeNet, which can be found on github. The paper descripting the training and validation of the model can be found here. Filenames are structured as follows: Data - [Modality]_[species Organ]_[resolution].tif Labels - [Modality]_[species Organ]_[resolution]_labels.tif Sub-volumes of larger dataset - [Modality]_[species Organ]_subvolume[dimensions in pixels].tif Manual labelling of blood vessels was carried out using Amira (2020.2, Thermo-Fisher, UK). Training data: opticalHREM_murineLiver_2.26x2.26x1.75um.tif: A high resolution episcopic microscopy (HREM) dataset, acquired in house by staining a healthy mouse liver with Eosin B and imaged using a standard HREM protocol. NB: 25% of this image volume was withheld from training, for use as test data. CT_murineTumour_20x20x20um.tif: X-ray microCT images of a microvascular cast, taken from a subcutaneous mouse model of colorectal cancer (acquired in house). NB: 25% of this image volume was withheld from training, for use as test data. RSOM_murineTumour_20x20um.tif: Raster-Scanning Optoacoustic Mesoscopy (RSOM) data from a subcutaneous tumour model (provided by Emma Brown, Bohndiek Group, University of Cambridge). The image data has undergone filtering to reduce the background ​(Brown et al., 2019)​. OCTA_humanRetina_24x24um.tif: retinal angiography data obtained using Optical Coherence Tomography Angiography (OCT-A) (provided by Dr Ranjan Rajendram, Moorfields Eye Hospital). Test data: MRI_porcineLiver_0.9x0.9x5mm.tif: T1-weighted Balanced Turbo Field Echo Magnetic Resonance Imaging (MRI) data from a machine-perfused porcine liver, acquired in-house. Test Data MFHREM_murineTumourLectin_2.76x2.76x2.61um.tif: a subcutaneous colorectal tumour mouse model was imaged in house using Multi-fluorescence HREM in house, with Dylight 647 conjugated lectin staining the vasculature ​(Walsh et al., 2021)​. The image data has been processed using an asymmetric deconvolution algorithm described by ​Walsh et al., 2020​. NB: A sub-volume of 480x480x640 voxels was manually labelled (MFHREM_murineTumourLectin_subvolume480x480x640.tif). MFHREM_murineBrainLectin_0.85x0.85x0.86um.tif: an MF-HREM image of the cortex of a mouse brain, stained with Dylight-647 conjugated lectin, was acquired in house ​(Walsh et al., 2021)​. The image data has been downsampled and processed using an asymmetric deconvolution algorithm described by ​Walsh et al., 2020​. NB: A sub-volume of 1000x1000x99 voxels was manually labelled. This sub-volume is provided at full resolution and without preprocessing (MFHREM_murineBrainLectin_subvol_0.57x0.57x0.86um.tif). 2Photon_murineOlfactoryBulbLectin_0.2x0.46x5.2um.tif: two-photon data of mouse olfactory bulb blood vessels, labelled with sulforhodamine 101, was kindly provided by Yuxin Zhang at the Sensory Circuits and Neurotechnology Lab, the Francis Crick Institute ​(Bosch et al., 2022)​. NB: A sub-volume of 500x500x79 voxel was manually labelled (2Photon_murineOlfactoryBulbLectin_subvolume500x500x79.tif). References: ​​Bosch, C., Ackels, T., Pacureanu, A., Zhang, Y., Peddie, C. J., Berning, M., Rzepka, N., Zdora, M. C., Whiteley, I., Storm, M., Bonnin, A., Rau, C., Margrie, T., Collinson, L., & Schaefer, A. T. (2022). Functional and multiscale 3D structural investigation of brain tissue through correlative in vivo physiology, synchrotron microtomography and volume electron microscopy. Nature Communications 2022 13:1, 13(1), 1–16. https://doi.org/10.1038/s41467-022-30199-6 ​Brown, E., Brunker, J., & Bohndiek, S. E. (2019). Photoacoustic imaging as a tool to probe the tumour microenvironment. DMM Disease Models and Mechanisms, 12(7). https://doi.org/10.1242/DMM.039636 ​Walsh, C., Holroyd, N. A., Finnerty, E., Ryan, S. G., Sweeney, P. W., Shipley, R. J., & Walker-Samuel, S. (2021). Multifluorescence High-Resolution Episcopic Microscopy for 3D Imaging of Adult Murine Organs. Advanced Photonics Research, 2(10), 2100110. https://doi.org/10.1002/ADPR.202100110 ​Walsh, C., Holroyd, N., Shipley, R., & Walker-Samuel, S. (2020). Asymmetric Point Spread Function Estimation and Deconvolution for Serial-Sectioning Block-Face Imaging. Communications in Computer and Information Science, 1248 CCIS, 235–249. https://doi.org/10.1007/978-3-030-52791-4_19 ​

Collective behaviour such as the flocks of birds and schools of fish is inspired by computer-based systems and is widely used in agents’ formation. The human could easily recognise these behaviours; however, it is hard for a computer system to recognise these behaviours. Since humans could easily recognise these behaviours, ground truth data on human perception of collective behaviour could enable machine learning methods to mimic this human perception. Hence ground truth data has been collected from human perception of collective behaviour recognition by running an online survey. Specific collective motions considered in this online survey include 16 structured and unstructured behaviours. The defined structured collective motions include boids’ movements with an identifiable embedded pattern. Unstructured collective motions consist of random movement of boids with no patterns. The participants are from diverse levels of knowledge, all over the world, and are over 18 years old. Each question contains a short video (around 10 seconds), captured from one of the 16 simulated movements. The videos are shown in a randomized order to the participants. Then they were asked to label each structured motion of boids as ‘flocking’, ‘aligned’, or ‘grouped’ and others as ‘not flocking’, ‘not aligned’, or ‘not grouped’. By averaging human perceptions, three binary labelled datasets of these motions are created. The data could be trained by machine learning methods, which enabled them to automatically recognise collective behaviour.

AI in Semi-supervised Learning Market Outlook

According to our latest research, the AI in Semi-supervised Learning market size reached USD 1.82 billion in 2024 globally, driven by rapid advancements in artificial intelligence and machine learning applications across diverse industries. The market is expected to expand at a robust CAGR of 28.1% from 2025 to 2033, reaching a projected value of USD 17.17 billion by 2033. This exponential growth is primarily fueled by the increasing need for efficient data labeling, the proliferation of unstructured data, and the growing adoption of AI-driven solutions in both large enterprises and small and medium businesses. As per the latest research, the surging demand for automation, accuracy, and cost-efficiency in data processing is significantly accelerating the adoption of semi-supervised learning models worldwide.

One of the most significant growth factors for the AI in Semi-supervised Learning market is the explosive increase in data generation across industries such as healthcare, finance, retail, and automotive. Organizations are continually collecting vast amounts of structured and unstructured data, but the process of labeling this data for supervised learning remains time-consuming and expensive. Semi-supervised learning offers a compelling solution by leveraging small amounts of labeled data alongside large volumes of unlabeled data, thus reducing the dependency on extensive manual annotation. This approach not only accelerates the deployment of AI models but also enhances their accuracy and scalability, making it highly attractive for enterprises seeking to maximize the value of their data assets while minimizing operational costs.

Another critical driver propelling the growth of the AI in Semi-supervised Learning market is the increasing sophistication of AI algorithms and the integration of advanced technologies such as deep learning, natural language processing, and computer vision. These advancements have enabled semi-supervised learning models to achieve remarkable performance in complex tasks like image and speech recognition, medical diagnostics, and fraud detection. The ability to process and interpret vast datasets with minimal supervision is particularly valuable in sectors where labeled data is scarce or expensive to obtain. Furthermore, the ongoing investments in research and development by leading technology companies and academic institutions are fostering innovation, resulting in more robust and scalable semi-supervised learning frameworks that can be seamlessly integrated into enterprise workflows.

The proliferation of cloud computing and the increasing adoption of hybrid and multi-cloud environments are also contributing significantly to the expansion of the AI in Semi-supervised Learning market. Cloud-based deployment offers unparalleled scalability, flexibility, and cost-efficiency, allowing organizations of all sizes to access cutting-edge AI tools and infrastructure without the need for substantial upfront investments. This democratization of AI technology is empowering small and medium enterprises to leverage semi-supervised learning for competitive advantage, driving widespread adoption across regions and industries. Additionally, the emergence of AI-as-a-Service (AIaaS) platforms is further simplifying the integration and management of semi-supervised learning models, enabling businesses to accelerate their digital transformation initiatives and unlock new growth opportunities.

From a regional perspective, North America currently dominates the AI in Semi-supervised Learning market, accounting for the largest share in 2024, followed closely by Europe and Asia Pacific. The strong presence of leading AI vendors, robust technological infrastructure, and high investments in AI research and development are key factors driving market growth in these regions. Asia Pacific is expected to witness the fastest CAGR during the forecast period, fueled by rapid digitalization, expanding IT infrastructure, and increasing government initiatives to promote AI adoption. Meanwhile, Latin America and the Middle East & Africa are also showing promising growth potential, supported by rising awareness of AI benefits and growing investments in digital transformation projects across various sectors.

Component Analysis

The component segment of the AI in Semi-supervised Learning market is divided into software, hardware, and services, each playing a pivotal role in the adoption and implementation of semi-s

The Digit Recognizer dataset is a classic benchmark used in the field of computer vision and machine learning for handwritten digit classification. It is derived from the well-known MNIST dataset, which consists of grayscale images of handwritten digits (0 through 9). Each image is a 28x28 pixel square, flattened into a 784-dimensional vector, and labeled with the corresponding digit.

This dataset serves as an excellent starting point for building and evaluating classification algorithms, particularly convolutional neural networks (CNNs). Due to its relatively small size and well-structured format, it allows for rapid experimentation and prototyping of models. Furthermore, the consistent quality of the images and the clear labeling make it a standard benchmark for comparing different machine learning approaches.

In this analysis, we will explore the dataset’s structure, perform data preprocessing, visualize example digits, and apply various machine learning models to evaluate their accuracy in classifying handwritten digits.

Telecom Data Labeling Market Outlook

According to our latest research, the global Telecom Data Labeling market size reached USD 1.42 billion in 2024, driven by the exponential growth in data generation, increasing adoption of AI and machine learning in telecom operations, and the rising complexity of communication networks. The market is forecasted to expand at a robust CAGR of 22.8% from 2025 to 2033, reaching an estimated USD 10.09 billion by 2033. This strong momentum is underpinned by the escalating demand for high-quality labeled datasets to power advanced analytics and automation in the telecom sector.

The growth trajectory of the Telecom Data Labeling market is fundamentally propelled by the surging data volumes generated by telecom networks worldwide. With the proliferation of 5G, IoT devices, and cloud-based services, telecom operators are inundated with massive streams of structured and unstructured data. Efficient data labeling is essential to transform raw data into actionable insights, fueling AI-driven solutions for network optimization, predictive maintenance, and fraud detection. Additionally, the mounting pressure on telecom companies to enhance customer experience and operational efficiency is prompting significant investments in data labeling infrastructure and services, further accelerating market expansion.

Another critical growth factor is the rapid evolution of artificial intelligence and machine learning applications within the telecommunications industry. AI-powered tools depend on vast quantities of accurately labeled data to deliver reliable predictions and automation. As telecom companies strive to automate network management, detect anomalies, and personalize user experiences, the demand for high-quality labeled datasets has surged. The emergence of advanced labeling techniques, including semi-automated and automated labeling methods, is enabling telecom enterprises to keep pace with the growing data complexity and volume, thus fostering faster and more scalable AI deployments.

Furthermore, regulatory compliance and data privacy concerns are shaping the landscape of the Telecom Data Labeling market. As governments worldwide tighten data protection regulations, telecom operators are compelled to ensure that data used for AI and analytics is accurately labeled and anonymized. This necessity is driving the adoption of robust data labeling solutions that not only facilitate compliance but also enhance data quality and integrity. The integration of secure, privacy-centric labeling platforms is becoming a competitive differentiator, especially in regions with stringent data governance frameworks. This trend is expected to persist, reinforcing the marketÂ’s upward trajectory.

AI-Powered Product Labeling is revolutionizing the telecom industry by providing more efficient and accurate data annotation processes. This technology leverages artificial intelligence to automate the labeling of large datasets, reducing the time and costs associated with manual labeling. By utilizing AI algorithms, telecom operators can ensure that their data is consistently labeled with high precision, which is crucial for training machine learning models. This advancement not only enhances the quality of labeled data but also accelerates the deployment of AI-driven solutions across various applications, such as network optimization and customer experience management. As AI-Powered Product Labeling continues to evolve, it is expected to play a pivotal role in the telecom sector's digital transformation journey, enabling operators to harness the full potential of their data assets.

From a regional perspective, Asia Pacific is emerging as a powerhouse in the Telecom Data Labeling market, fueled by rapid digitalization, expanding telecom infrastructure, and the early adoption of 5G technologies. North America remains a significant contributor, owing to its mature telecom ecosystem and high investments in AI research and development. Europe is also witnessing steady growth, driven by regulatory mandates and increasing focus on data-driven network management. Meanwhile, Latin America and the Middle East & Africa are gradually catching up, with investments in digital transformation and telecom modernization initiatives providing new growth avenues. These regional dynamics collectively underscore the global nature

Dataset Description: Human Faces and Objects Dataset (HFO-5000) The Human Faces and Objects Dataset (HFO-5000) is a curated collection of 5,000 images, categorized into three distinct classes: male faces (1,500), female faces (1,500), and objects (2,000). This dataset is designed for machine learning and computer vision applications, including image classification, face detection, and object recognition. The dataset provides high-quality, labeled images with a structured CSV file for seamless integration into deep learning pipelines.

Column Description: The dataset is accompanied by a CSV file that contains essential metadata for each image. The CSV file includes the following columns: file_name: The name of the image file (e.g., image_001.jpg). label: The category of the image, with three possible values: "male" (for male face images) "female" (for female face images) "object" (for images of various objects) file_path: The full or relative path to the image file within the dataset directory.

Uniqueness and Key Features: 1) Balanced Distribution: The dataset maintains an even distribution of human faces (male and female) to minimize bias in classification tasks. 2) Diverse Object Selection: The object category consists of a wide variety of items, ensuring robustness in distinguishing between human and non-human entities. 3) High-Quality Images: The dataset consists of clear and well-defined images, suitable for both training and testing AI models. 4) Structured Annotations: The CSV file simplifies dataset management and integration into machine learning workflows. 5) Potential Use Cases: This dataset can be used for tasks such as gender classification, facial recognition benchmarking, human-object differentiation, and transfer learning applications.

Conclusion: The HFO-5000 dataset provides a well-structured, diverse, and high-quality set of labeled images that can be used for various computer vision tasks. Its balanced distribution of human faces and objects ensures fairness in training AI models, making it a valuable resource for researchers and developers. By offering structured metadata and a wide range of images, this dataset facilitates advancements in deep learning applications related to facial recognition and object classification.

🧠 Dataset Title: Human-AI Preference Alignment (Inspired by Anthropic’s HH-RLHF) 📘 Overview: This dataset presents a curated collection of human-AI interaction samples designed to support cutting-edge research in Reinforcement Learning from Human Feedback (RLHF), ethical AI development, and model alignment. It follows the structure and spirit of the original hh-rlhf, making it a high-impact resource for fine-tuning and evaluating Large Language Models (LLMs).

Whether you're working on alignment, instruction-following behavior, safety, or human preference modeling, this dataset provides a strong foundation for experimentation and development.

🧩 What’s Inside? ✅ Thousands of preference-labeled response pairs, where annotators select the more aligned AI reply

✅ Multi-turn conversations between human prompts and assistant completions

✅ Designed for reward model training, RLHF pipelines, and supervised fine-tuning

✅ Structured in a way that supports both transformer-based and reinforcement learning models

✅ Covers a wide range of topics, from factual QA to ethical dilemmas and role-play

🎯 Use Cases: 🔹 Train reward models for instruction-following AI (e.g., InstructGPT, Claude, ChatGPT-style agents)

🔹 Evaluate LLM alignment with human values like helpfulness, harmlessness, and honesty (HHH)

🔹 Fine-tune open-source models (e.g., LLaMA, Mistral, Falcon, Gemma) using RLHF pipelines

🔹 Build preference-based datasets for safe and interpretable AI systems

🔹 Use in comparative learning tasks, conversational modeling, or safety benchmarking

🌍 Why This Dataset Matters: As AI systems become more capable, aligning their behavior with human ethical preferences becomes critically important. Human feedback is at the core of building AI that can reason, act safely, and respond meaningfully. This dataset contributes to that mission by offering high-quality, human-labeled data that reflects real-world human expectations in AI responses.

By enabling fine-tuning of models with reinforcement learning from actual human judgments, this dataset brings us one step closer to building trustworthy AI.

🧪 Inspirations & References: Anthropic’s HH-RLHF

OpenAI’s InstructGPT

Constitutional AI & Ethical Alignment techniques

Reward Modeling in Reinforcement Learning

📌 Tags / Keywords: AI Alignment • RLHF • Large Language Models • Reward Modeling • Preference Comparison • Ethical AI • Human Feedback • Open-source Fine-tuning

💬 Citation & Credit: If you use this dataset in your research, demos, or fine-tuning workflows, please cite the original HH-RLHF dataset and acknowledge this Kaggle version as an adapted resource for open-access experimentation.

This dataset contains metadata (title, abstract, date of publication, field, etc) for around 1 million academic articles. Each record contains additional information on the country of study and whether the article makes use of data. Machine learning tools were used to classify the country of study and data use.

Our data source of academic articles is the Semantic Scholar Open Research Corpus (S2ORC) (Lo et al. 2020). The corpus contains more than 130 million English language academic papers across multiple disciplines. The papers included in the Semantic Scholar corpus are gathered directly from publishers, from open archives such as arXiv or PubMed, and crawled from the internet.

We placed some restrictions on the articles to make them usable and relevant for our purposes. First, only articles with an abstract and parsed PDF or latex file are included in the analysis. The full text of the abstract is necessary to classify the country of study and whether the article uses data. The parsed PDF and latex file are important for extracting important information like the date of publication and field of study. This restriction eliminated a large number of articles in the original corpus. Around 30 million articles remain after keeping only articles with a parsable (i.e., suitable for digital processing) PDF, and around 26% of those 30 million are eliminated when removing articles without an abstract. Second, only articles from the year 2000 to 2020 were considered. This restriction eliminated an additional 9% of the remaining articles. Finally, articles from the following fields of study were excluded, as we aim to focus on fields that are likely to use data produced by countries’ national statistical system: Biology, Chemistry, Engineering, Physics, Materials Science, Environmental Science, Geology, History, Philosophy, Math, Computer Science, and Art. Fields that are included are: Economics, Political Science, Business, Sociology, Medicine, and Psychology. This third restriction eliminated around 34% of the remaining articles. From an initial corpus of 136 million articles, this resulted in a final corpus of around 10 million articles.

Due to the intensive computer resources required, a set of 1,037,748 articles were randomly selected from the 10 million articles in our restricted corpus as a convenience sample.

The empirical approach employed in this project utilizes text mining with Natural Language Processing (NLP). The goal of NLP is to extract structured information from raw, unstructured text. In this project, NLP is used to extract the country of study and whether the paper makes use of data. We will discuss each of these in turn.

To determine the country or countries of study in each academic article, two approaches are employed based on information found in the title, abstract, or topic fields. The first approach uses regular expression searches based on the presence of ISO3166 country names. A defined set of country names is compiled, and the presence of these names is checked in the relevant fields. This approach is transparent, widely used in social science research, and easily extended to other languages. However, there is a potential for exclusion errors if a country’s name is spelled non-standardly.

The second approach is based on Named Entity Recognition (NER), which uses machine learning to identify objects from text, utilizing the spaCy Python library. The Named Entity Recognition algorithm splits text into named entities, and NER is used in this project to identify countries of study in the academic articles. SpaCy supports multiple languages and has been trained on multiple spellings of countries, overcoming some of the limitations of the regular expression approach. If a country is identified by either the regular expression search or NER, it is linked to the article. Note that one article can be linked to more than one country.

The second task is to classify whether the paper uses data. A supervised machine learning approach is employed, where 3500 publications were first randomly selected and manually labeled by human raters using the Mechanical Turk service (Paszke et al. 2019).[1] To make sure the human raters had a similar and appropriate definition of data in mind, they were given the following instructions before seeing their first paper:

Each of these documents is an academic article. The goal of this study is to measure whether a specific academic article is using data and from which country the data came.

There are two classification tasks in this exercise:

1. identifying whether an academic article is using data from any country

2. Identifying from which country that data came.

For task 1, we are looking specifically at the use of data. Data is any information that has been collected, observed, generated or created to produce research findings. As an example, a study that reports findings or analysis using a survey data, uses data. Some clues to indicate that a study does use data includes whether a survey or census is described, a statistical model estimated, or a table or means or summary statistics is reported.

After an article is classified as using data, please note the type of data used. The options are population or business census, survey data, administrative data, geospatial data, private sector data, and other data. If no data is used, then mark "Not applicable". In cases where multiple data types are used, please click multiple options.[2]

For task 2, we are looking at the country or countries that are studied in the article. In some cases, no country may be applicable. For instance, if the research is theoretical and has no specific country application. In some cases, the research article may involve multiple countries. In these cases, select all countries that are discussed in the paper.

We expect between 10 and 35 percent of all articles to use data.

The median amount of time that a worker spent on an article, measured as the time between when the article was accepted to be classified by the worker and when the classification was submitted was 25.4 minutes. If human raters were exclusively used rather than machine learning tools, then the corpus of 1,037,748 articles examined in this study would take around 50 years of human work time to review at a cost of $3,113,244, which assumes a cost of $3 per article as was paid to MTurk workers.

A model is next trained on the 3,500 labelled articles. We use a distilled version of the BERT (bidirectional Encoder Representations for transformers) model to encode raw text into a numeric format suitable for predictions (Devlin et al. (2018)). BERT is pre-trained on a large corpus comprising the Toronto Book Corpus and Wikipedia. The distilled version (DistilBERT) is a compressed model that is 60% the size of BERT and retains 97% of the language understanding capabilities and is 60% faster (Sanh, Debut, Chaumond, Wolf 2019). We use PyTorch to produce a model to classify articles based on the labeled data. Of the 3,500 articles that were hand coded by the MTurk workers, 900 are fed to the machine learning model. 900 articles were selected because of computational limitations in training the NLP model. A classification of “uses data” was assigned if the model predicted an article used data with at least 90% confidence.

The performance of the models classifying articles to countries and as using data or not can be compared to the classification by the human raters. We consider the human raters as giving us the ground truth. This may underestimate the model performance if the workers at times got the allocation wrong in a way that would not apply to the model. For instance, a human rater could mistake the Republic of Korea for the Democratic People’s Republic of Korea. If both humans and the model perform the same kind of errors, then the performance reported here will be overestimated.

The model was able to predict whether an article made use of data with 87% accuracy evaluated on the set of articles held out of the model training. The correlation between the number of articles written about each country using data estimated under the two approaches is given in the figure below. The number of articles represents an aggregate total of

The provided dataset comprises 43 instances of temporal bone volume CT scans. The scans were performed on human cadaveric specimen with a resulting isotropic voxel size of \(99 \times 99 \times 99 \, \, \mathrm{\mu m}^3\). Voxel-wise image labels of the fluid space of the bony labyrinth, subdivided in the three semantic classes cochlear volume, vestibular volume and semicircular canal volume are provided. In addition, each dataset contains JSON-like descriptor data defining the voxel coordinates of the anatomical landmarks: (1) apex of the cochlea, (2) oval window and (3) round window. The dataset can be used to train and evaluate algorithmic machine learning models for automated innear ear analysis in the context of the supervised learning paradigm.

Usage Notes

The datasets are formatted in the HDF5 format developed by the HDF5 Group. We utilized and thus recommend the usage of Python bindings pyHDF to handle the datasets.

The flat-panel volume CT raw data, labels and landmarks are saved in the HDF5-internal file structure using the respective group and datasets:

raw/raw-0
label/label-0
landmark/landmark-0
landmark/landmark-1
landmark/landmark-2

Array raw and label data can be read from the file by indexing into an opened h5py file handle, for example as numpy.ndarray. Further metadata is contained in the attribute dictionaries of the raw and label datasets.

Landmark coordinate data is available as an attribute dict and contains the coordinate system (LPS or RAS), IJK voxel coordinates and label information. The helicotrema or cochlea top is globally saved in landmark 0, the oval window in landmark 1 and the round window in landmark 2. Read as a Python dictionary, exemplary landmark information for a dataset may reads as follows:

{'coordsys': 'LPS',
 'id': 1,
 'ijk_position': array([181, 188, 100]),
 'label': 'CochleaTop',
 'orientation': array([-1., -0., -0., -0., -1., -0., 0., 0., 1.]),
 'xyz_position': array([ 44.21109689, -139.38058589, -183.48249736])}

{'coordsys': 'LPS',
 'id': 2,
 'ijk_position': array([222, 182, 145]),
 'label': 'OvalWindow',
 'orientation': array([-1., -0., -0., -0., -1., -0., 0., 0., 1.]),
 'xyz_position': array([ 48.27890112, -139.95991131, -179.04103763])}

{'coordsys': 'LPS',
 'id': 3,
 'ijk_position': array([223, 209, 147]),
 'label': 'RoundWindow',
 'orientation': array([-1., -0., -0., -0., -1., -0., 0., 0., 1.]),
 'xyz_position': array([ 48.33120126, -137.27135678, -178.8665465 ])}

The MAAD dataset represents a comprehensive collection of Arabic news articles that may be employed across a diverse array of Arabic Natural Language Processing (NLP) applications, including but not limited to classification, text generation, summarization, and various other tasks. The dataset was diligently assembled through the application of specifically designed Python scripts that targeted six prominent news platforms: Al Jazeera, BBC Arabic, Youm7, Russia Today, and Al Ummah, in conjunction with regional and local media outlets, ultimately resulting in a total of 602,792 articles. This dataset exhibits a total word count of 29,371,439, with the number of unique words totaling 296,518; the average word length has been determined to be 6.36 characters, while the mean article length is calculated at 736.09 characters. This extensive dataset is categorized into ten distinct classifications: Political, Economic, Cultural, Arts, Sports, Health, Technology, Community, Incidents, and Local. The data fields are categorized into five distinct types: Title, Article, Summary, Category, and Published_ Date. The MAAD dataset is structured into six files, each named after the corresponding news outlets from which the data was sourced; within each directory, text files are provided, containing the number of categories represented in a single file, formatted in txt to accommodate all news articles. This dataset serves as an expansive standard resource designed for utilization within the context of our research endeavors.

The Data Annotation Service Market size was valued at USD 1.89 Billion in 2024 and is projected to reach USD 10.07 Billion by 2032, growing at a CAGR of 23% from 2026 to 2032.Global Data Annotation Service Market DriversThe data annotation service market is experiencing robust growth, propelled by the ever-increasing demand for high-quality, labeled data to train sophisticated artificial intelligence (AI) and machine learning (ML) models. As AI continues to permeate various industries, the need for accurate and diverse datasets becomes paramount, making data annotation a critical component of successful AI development. This article explores the key drivers fueling the expansion of the data annotation service market.Rising Demand for Artificial Intelligence (AI) and Machine Learning (ML) Applications: One of the most influential drivers of the data annotation service market is the surging adoption of artificial intelligence (AI) and machine learning (ML) across industries. Data annotation plays a critical role in training AI algorithms to recognize, categorize, and interpret real-world data accurately. From autonomous vehicles to medical diagnostics, annotated datasets are essential for improving model accuracy and performance. As enterprises expand their AI initiatives, they increasingly rely on professional annotation services to handle large, complex, and diverse datasets. This trend is expected to accelerate as AI continues to penetrate industries such as healthcare, finance, automotive, and retail, driving steady market growth.Expansion of Autonomous Vehicle Development: The growing focus on autonomous vehicle technology is a major catalyst for the data annotation service industry. Self-driving cars require immense volumes of labeled image and video data to identify pedestrians, road signs, vehicles, and lane markings with precision.

Curated image privacy data sets

In the article, we trained and evaluated models on the Image Privacy Dataset (IPD) and the PrivacyAlert dataset. The datasets are originally provided by other sources and have been re-organised and curated for this work.

Our curation organises the datasets in a common structure. We updated the annotations and labelled the splits of the data in the annotation file. This avoids having separated folders of images for each data split (training, validation, testing) and allows a flexible handling of new splits, e.g. created with a stratified K-Fold cross-validation procedure. As for the original datasets (PicAlert and PrivacyAlert), we provide the link to the images in bash scripts to download the images. Another bash script re-organises the images in sub-folders with maximum 1000 images in each folder.

Both datasets refer to images publicly available on Flickr. These images have a large variety of content, including sensitive content, seminude people, vehicle plates, documents, private events. Images were annotated with a binary label denoting if the content was deemed to be public or private. As the images are publicly available, their label is mostly public. These datasets have therefore a high imbalance towards the public class. Note that IPD combines two other existing datasets, PicAlert and part of VISPR, to increase the number of private images already limited in PicAlert. Further details in our corresponding https://doi.org/10.48550/arXiv.2503.12464" target="_blank" rel="noopener">publication.

List of datasets and their original source:

• PicAlert [Images occupy 2.4 GB]
• VISPR [Images occupy 49.7 GB]
• PrivacyAlert [Images occupy 1 GB]

Notes:

• For PicAlert and PrivacyAlert, only urls to the original locations in Flickr are available in the Zenodo record
• Collector and authors of the PrivacyAlert dataset selected the images from Flickr under Public Domain license
• Owners of the photos on Flick could have removed the photos from the social media platform
• Running the bash scripts to download the images can incur in the "429 Too Many Requests" status code

Pre-computed visual entitities

Some of the models run their pipeline end-to-end with the images as input, whereas other models require different or additional inputs. These inputs include the pre-computed visual entities (scene types and object types) represented in a graph format, e.g. for a Graph Neural Network. Re-using these pre-computed visual entities allows other researcher to build new models based on these features while avoiding re-computing the same on their own or for each epoch during the training of a model (faster training).

For each image of each dataset, namely PrivacyAlert, PicAlert, and VISPR, we provide the predicted scene probabilities as a .csv file , the detected objects as a .json file in COCO data format, and the node features (visual entities already organised in graph format with their features) as a .json file. For consistency, all the files are already organised in batches following the structure of the images in the datasets folder. For each dataset, we also provide the pre-computed adjacency matrix for the graph data.

Note: IPD is based on PicAlert and VISPR and therefore IPD refers to the scene probabilities and object detections of the other two datasets. Both PicAlert and VISPR must be downloaded and prepared to use IPD for training and testing.

Further details on downloading and organising data can be found in our GitHub repository: https://github.com/graphnex/privacy-from-visual-entities (see ARTIFACT-EVALUATION.md#pre-computed-visual-entitities-)

Enquiries, questions and comments

If you have any enquiries, question, or comments, or you would like to file a bug report or a feature request, use the issue tracker of our GitHub repository.

Market Size and Growth: The global Structured Product Labeling Management market is valued at XXX million in 2025 and is projected to reach XXX million by 2033, exhibiting a CAGR of XX%. This growth is attributed to the increasing complexity of regulatory requirements for pharmaceutical products, the rise in outsourcing of labeling processes by biopharmaceutical companies, and the adoption of digital solutions to streamline labeling workflows. Drivers and Trends: Key drivers include the increasing demand for accurate and compliant labeling to ensure patient safety, the adoption of personalized medicine and companion diagnostics, and the need for faster time-to-market for new products. Moreover, the trend towards cloud-based and mobile solutions, as well as the integration of artificial intelligence (AI) and machine learning (ML) technologies, is expected to further drive market growth. However, factors such as data security concerns and the need for trained workforce may restrain market expansion.

Data and metadata used in "Machine learning reveals the waggle drift’s role in the honey bee dance communication system"

All timestamps are given in ISO 8601 format.

The following files are included:

Berlin2019_waggle_phases.csv, Berlin2021_waggle_phases.csv

Automatic individual detections of waggle phases during our recording periods in 2019 and 2021.

timestamp: Date and time of the detection.

cam_id: Camera ID (0: left side of the hive, 1: right side of the hive).

x_median, y_median: Median position of the bee during the waggle phase (for 2019 given in millimeters after applying a homography, for 2021 in the original image coordinates).

waggle_angle: Body orientation of the bee during the waggle phase in radians (0: oriented to the right, PI / 4: oriented upwards).

Berlin2019_dances.csv

Automatic detections of dance behavior during our recording period in 2019.

dancer_id: Unique ID of the individual bee.

dance_id: Unique ID of the dance.

ts_from, ts_to: Date and time of the beginning and end of the dance.

cam_id: Camera ID (0: left side of the hive, 1: right side of the hive).

median_x, median_y: Median position of the individual during the dance.

feeder_cam_id: ID of the feeder that the bee was detected at prior to the dance.

Berlin2019_followers.csv

Automatic detections of attendance and following behavior, corresponding to the dances in Berlin2019_dances.csv.

dance_id: Unique ID of the dance being attended or followed.

follower_id: Unique ID of the individual attending or following the dance.

ts_from, ts_to: Date and time of the beginning and end of the interaction.

label: “attendance” or “follower”

cam_id: Camera ID (0: left side of the hive, 1: right side of the hive).

Berlin2019_dances_with_manually_verified_times.csv

A sample of dances from Berlin2019_dances.csv where the exact timestamps have been manually verified to correspond to the beginning of the first and last waggle phase down to a precision of ca. 166 ms (video material was recorded at 6 FPS).

dance_id: Unique ID of the dance.

dancer_id: Unique ID of the dancing individual.

cam_id: Camera ID (0: left side of the hive, 1: right side of the hive).

feeder_cam_id: ID of the feeder that the bee was detected at prior to the dance.

dance_start, dance_end: Manually verified date and times of the beginning and end of the dance.

Berlin2019_dance_classifier_labels.csv

Manually annotated waggle phases or following behavior for our recording season in 2019 that was used to train the dancing and following classifier. Can be merged with the supplied individual detections.

timestamp: Timestamp of the individual frame the behavior was observed in.

frame_id: Unique ID of the video frame the behavior was observed in.

bee_id: Unique ID of the individual bee.

label: One of “nothing”, “waggle”, “follower”

Berlin2019_dance_classifier_unlabeled.csv

Additional unlabeled samples of timestamp and individual ID with the same format as Berlin2019_dance_classifier_labels.csv, but without a label. The data points have been sampled close to detections of our waggle phase classifier, so behaviors related to the waggle dance are likely overrepresented in that sample.

Berlin2021_waggle_phase_classifier_labels.csv

Manually annotated detections of our waggle phase detector (bb_wdd2) that were used to train the neural network filter (bb_wdd_filter) for the 2021 data.

detection_id: Unique ID of the waggle phase.

label: One of “waggle”, “activating”, “ventilating”, “trembling”, “other”. Where “waggle” denoted a waggle phase, “activating” is the shaking signal, “ventilating” is a bee fanning her wings. “trembling” denotes a tremble dance, but the distinction from the “other” class was often not clear, so “trembling” was merged into “other” for training.

orientation: The body orientation of the bee that triggered the detection in radians (0: facing to the right, PI /4: facing up).

metadata_path: Path to the individual detection in the same directory structure as created by the waggle dance detector.

Berlin2021_waggle_phase_classifier_ground_truth.zip

The output of the waggle dance detector (bb_wdd2) that corresponds to Berlin2021_waggle_phase_classifier_labels.csv and is used for training. The archive includes a directory structure as output by the bb_wdd2 and each directory includes the original image sequence that triggered the detection in an archive and the corresponding metadata. The training code supplied in bb_wdd_filter directly works with this directory structure.

Berlin2019_tracks.zip

Detections and tracks from the recording season in 2019 as produced by our tracking system. As the full data is several terabytes in size, we include the subset of our data here that is relevant for our publication which comprises over 46 million detections. We included tracks for all detected behaviors (dancing, following, attending) including one minute before and after the behavior. We also included all tracks that correspond to the labeled and unlabeled data that was used to train the dance classifier including 30 seconds before and after the data used for training. We grouped the exported data by date to make the handling easier, but to efficiently work with the data, we recommend importing it into an indexable database.

The individual files contain the following columns:

cam_id: Camera ID (0: left side of the hive, 1: right side of the hive).

timestamp: Date and time of the detection.

frame_id: Unique ID of the video frame of the recording from which the detection was extracted.

track_id: Unique ID of an individual track (short motion path from one individual). For longer tracks, the detections can be linked based on the bee_id.

bee_id: Unique ID of the individual bee.

bee_id_confidence: Confidence between 0 and 1 that the bee_id is correct as output by our tracking system.

x_pos_hive, y_pos_hive: Spatial position of the bee in the hive on the side indicated by cam_id. Given in millimeters after applying a homography on the video material.

orientation_hive: Orientation of the bees’ thorax in the hive in radians (0: oriented to the right, PI / 4: oriented upwards).

Berlin2019_feeder_experiment_log.csv

Experiment log for our feeder experiments in 2019.

date: Date given in the format year-month-day.

feeder_cam_id: Numeric ID of the feeder.

coordinates: Longitude and latitude of the feeder. For feeders 1 and 2 this is only given once and held constant. Feeder 3 had varying locations.

time_opened, time_closed: Date and time when the feeder was set up or closed again. sucrose_solution: Concentration of the sucrose solution given as sugar:water (in terms of weight). On days where feeder 3 was open, the other two feeders offered water without sugar.

Software used to acquire and analyze the data:

bb_pipeline: Tag localization and decoding pipeline

bb_pipeline_models: Pretrained localizer and decoder models for bb_pipeline

bb_binary: Raw detection data storage format

bb_irflash: IR flash system schematics and arduino code

bb_imgacquisition: Recording and network storage

bb_behavior: Database interaction and data (pre)processing, feature extraction

bb_tracking: Tracking of bee detections over time

bb_wdd2: Automatic detection and decoding of honey bee waggle dances

bb_wdd_filter: Machine learning model to improve the accuracy of the waggle dance detector

bb_dance_networks: Detection of dancing and following behavior from trajectories

3DIFICE: 3-dimensional Damage Imposed on Frame structures for Investigating Computer vision-based Evaluation methods This dataset contains 1,396 synthetic images and label maps with various types of earthquake damage imposed on reinforced concrete frame structures. Damage includes: cracking, spalling, exposed transverse rebar, and exposed longitudinal rebar. Each image has an associated label map that can be used for training machine learning algorithms to recognize the various types of damage.