Data Labeling Software Market Outlook

In 2023, the global market size for data labeling software was valued at approximately USD 1.2 billion and is projected to reach USD 6.5 billion by 2032, with a CAGR of 21% during the forecast period. The primary growth factor driving this market is the increasing adoption of artificial intelligence (AI) and machine learning (ML) technologies across various industry verticals, necessitating high-quality labeled data for model training and validation.

The surge in AI and ML applications is a significant growth driver for the data labeling software market. As businesses increasingly harness these advanced technologies to gain insights, optimize operations, and innovate products and services, the demand for accurately labeled data has skyrocketed. This trend is particularly pronounced in sectors such as healthcare, automotive, and finance, where AI and ML applications are critical for advancements like predictive analytics, autonomous driving, and fraud detection. The growing reliance on AI and ML is propelling the market forward, as labeled data forms the backbone of effective AI model development.

Another crucial growth factor is the proliferation of big data. With the explosion of data generated from various sources, including social media, IoT devices, and enterprise systems, organizations are seeking efficient ways to manage and utilize this vast amount of information. Data labeling software enables companies to systematically organize and annotate large datasets, making them usable for AI and ML applications. The ability to handle diverse data types, including text, images, and audio, further amplifies the demand for these solutions, facilitating more comprehensive data analysis and better decision-making.

The increasing emphasis on data privacy and security is also driving the growth of the data labeling software market. With stringent regulations such as GDPR and CCPA coming into play, companies are under pressure to ensure that their data handling practices comply with legal standards. Data labeling software helps in anonymizing and protecting sensitive information during the labeling process, thus providing a layer of security and compliance. This has become particularly important as data breaches and cyber threats continue to rise, making secure data management a top priority for organizations worldwide.

Regionally, North America holds a significant share of the data labeling software market due to early adoption of AI and ML technologies, substantial investments in tech startups, and advanced IT infrastructure. However, the Asia Pacific region is expected to witness the highest growth rate during the forecast period. This growth is driven by the rapid digital transformation in countries like China and India, increasing investments in AI research, and the expansion of IT services. Europe and Latin America also present substantial growth opportunities, supported by technological advancements and increasing regulatory compliance needs.

Component Analysis

The data labeling software market can be segmented by component into software and services. The software segment encompasses various platforms and tools designed to label data efficiently. These software solutions offer features such as automation, integration with other AI tools, and scalability, which are critical for handling large datasets. The growing demand for automated data labeling solutions is a significant trend in this segment, driven by the need for faster and more accurate data annotation processes.

In contrast, the services segment includes human-in-the-loop solutions, consulting, and managed services. These services are essential for ensuring the quality and accuracy of labeled data, especially for complex tasks that require human judgment. Companies often turn to service providers for their expertise in specific domains, such as healthcare or automotive, where domain knowledge is crucial for effective data labeling. The services segment is also seeing growth due to the increasing need for customized solutions tailored to specific business requirements.

Moreover, hybrid approaches that combine software and human expertise are gaining traction. These solutions leverage the scalability and speed of automated software while incorporating human oversight for quality assurance. This combination is particularly useful in scenarios where data quality is paramount, such as in medical imaging or autonomous vehicle training. The hybrid model is expected to grow as companies seek to balance efficiency with accuracy in their

Data Annotation And Labeling Market Size And Forecast

Data Annotation And Labeling Market size was valued to be USD 1080.8 Million in the year 2023 and it is expected to reach USD 8851.05 Million in 2031, growing at a CAGR of 35.10% from 2024 to 2031.

Data Annotation And Labeling Market Drivers

Increased Adoption of Artificial Intelligence (AI) and Machine Learning (ML): The demand for large volumes of high-quality labeled data to effectively train these systems is being driven by the widespread adoption of AI and ML technologies across various industries, thereby fueling the growth of the Data Annotation And Labeling Market.

Advancements in Computer Vision and Natural Language Processing: A need for annotated and labeled data to develop and enhance AI models capable of understanding and interpreting visual and textual data accurately is created by the rapid progress in fields such as computer vision and natural language processing.

Growth of Cloud Computing and Big Data: The adoption of AI and ML solutions has been facilitated by the rise of cloud computing and the availability of massive amounts of data, leading to an increased demand for data annotation and labeling services to organize and prepare this data for analysis and model training.

As per Cognitive Market Research's latest published report, the Global Data Collection and Labeling market size was USD 2.41 Billion in 2022 and it is forecasted to reach USD 18.60 Billion by 2030. Data Collection and Labeling Industry's Compound Annual Growth Rate will be 29.1% from 2023 to 2030. Key Dynamics of Data Collection And Labeling Market

Key Drivers of Data Collection And Labeling Market

Surge in AI and Machine Learning Adoption: The increasing integration of AI across various industries has led to a notable rise in the demand for high-quality labeled datasets. Precise data labeling is essential for training machine learning models, particularly in fields such as autonomous vehicles, healthcare diagnostics, and facial recognition.

Proliferation of Unstructured Data: With the surge of images, videos, and audio data generated from digital platforms, businesses are in need of structured labeling services to transform raw data into usable datasets. This trend is propelling the growth of data annotation services, especially for applications in natural language processing and computer vision.

Rising Use in Healthcare and Retail: Data labeling plays a vital role in applications such as medical imaging, drug discovery, and e-commerce personalization. Industries like healthcare and retail are allocating resources towards labeled datasets to enhance AI-driven diagnostics, recommendation systems, and predictive analytics, thereby increasing market demand.

Key Restrains for Data Collection And Labeling Market

High Cost and Time-Intensive Process: The process of manual data labeling is both labor-intensive and costly, particularly for intricate projects that necessitate expert annotators. This can pose a challenge for small businesses or startups that operate with limited budgets and stringent development timelines.

Data Privacy and Compliance Challenges: Managing sensitive information, including personal photographs, biometric data, or patient records, raises significant concerns regarding security and regulatory compliance. Ensuring compliance with GDPR, HIPAA, or other data protection regulations complicates the data labeling process.

Lack of Skilled Workforce: The industry is experiencing a shortage of qualified data annotators, especially in specialized areas such as radiology or autonomous systems. The inconsistency in labeling quality due to insufficient domain expertise can adversely affect the accuracy and reliability of AI models.

Key Trends in Data Collection And Labelingl Market

Emergence of Automated and Semi-Automated Labeling Tools: Companies are progressively embracing AI-driven labeling tools to minimize manual labor. Innovations such as active learning, auto-labeling, and transfer learning are enhancing efficiency and accelerating the data preparation workflow.

Expansion of Crowdsourcing Platforms: Crowdsourced data labeling via platforms like Amazon Mechanical Turk is gaining traction as a favored approach. It facilitates quicker turnaround times at reduced costs by utilizing a global workforce, particularly for tasks involving image classification, sentiment analysis, and object detection.

Transition Towards Industry-Specific Labeling Solutions: Providers are creating domain-specific labeling platforms customized for sectors such as agriculture, autonomous vehicles, or legal technology. These specialized tools enhance accuracy, shorten time-to-market, and cater to the specific requirements of vertical AI applications. What is Data Collection and Labeling?

Data collection and labeling is the process of gathering and organizing data and adding metadata to it for better analysis and understanding. This process is critical in machine learning and artificial intelligence, as it provides the foundation for training algorithms that can identify patterns and make predictions. Data collection involves gathering raw data from various sources, including sensors, databases, websites, and other forms of digital media. The collected data may be unstructured or structured, and it may be in different formats, such as text, images, videos, or audio.

Context: This dataset consists of images from four main categories: Nature, Fashion, Food, and Animals. However, each category includes a mixture of images that overlap with the other categories. For instance, the Nature category contains not only natural landscapes but also fashion items related to nature, food items grown in nature, and animals in natural settings. Similarly, each of the other categories follows the same pattern, making this dataset unique and versatile for complex classification tasks.

Content: The dataset includes images distributed across four main classes:

Nature: Images of landscapes, plants, animals in natural habitats, nature-inspired fashion, and natural food items. Fashion: Clothing, accessories, and fashion styles, including food and animals depicted in fashion-related contexts. Food: Different types of food, food items in fashion or nature settings, and animals involved with food production. Animals: Images of various animal species, animals in natural environments, animals used in fashion, and animals related to food. Each class contains a variety of images that are unclassified within their specific category, making the dataset ideal for multi-class classification, hierarchical categorization, and data labeling tasks.

Usage: This dataset is designed for:

Multi-class Image Classification: Models that need to classify images into one of four broad categories. Hierarchical Classification: Using sub-categories within each class (e.g., Fashion-Nature, Food-Animals) to refine classifications. Data Labeling Tasks: Researchers or developers can label and organize the images into more specific sub-categories for improved dataset structure.

FSDKaggle2019 is an audio dataset containing 29,266 audio files annotated with 80 labels of the AudioSet Ontology. FSDKaggle2019 has been used for the DCASE Challenge 2019 Task 2, which was run as a Kaggle competition titled Freesound Audio Tagging 2019.

Citation

If you use the FSDKaggle2019 dataset or part of it, please cite our DCASE 2019 paper:

Eduardo Fonseca, Manoj Plakal, Frederic Font, Daniel P. W. Ellis, Xavier Serra. "Audio tagging with noisy labels and minimal supervision". Proceedings of the DCASE 2019 Workshop, NYC, US (2019)

You can also consider citing our ISMIR 2017 paper, which describes how we gathered the manual annotations included in FSDKaggle2019.

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

Data curators

Eduardo Fonseca, Manoj Plakal, Xavier Favory, Jordi Pons

Contact

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

ABOUT FSDKaggle2019

Freesound Dataset Kaggle 2019 (or FSDKaggle2019 for short) is an audio dataset containing 29,266 audio files annotated with 80 labels of the AudioSet Ontology [1]. FSDKaggle2019 has been used for the Task 2 of the Detection and Classification of Acoustic Scenes and Events (DCASE) Challenge 2019. Please visit the DCASE2019 Challenge Task 2 website for more information. This Task was hosted on the Kaggle platform as a competition titled Freesound Audio Tagging 2019. It was organized by researchers from the Music Technology Group (MTG) of Universitat Pompeu Fabra (UPF), and from Sound Understanding team at Google AI Perception. The competition intended to provide insight towards the development of broadly-applicable sound event classifiers able to cope with label noise and minimal supervision conditions.

FSDKaggle2019 employs audio clips from the following sources:

Freesound Dataset (FSD): a dataset being collected at the MTG-UPF based on Freesound content organized with the AudioSet Ontology

The soundtracks of a pool of Flickr videos taken from the Yahoo Flickr Creative Commons 100M dataset (YFCC)

The audio data is labeled using a vocabulary of 80 labels from Google’s AudioSet Ontology [1], covering diverse topics: Guitar and other Musical Instruments, Percussion, Water, Digestive, Respiratory sounds, Human voice, Human locomotion, Hands, Human group actions, Insect, Domestic animals, Glass, Liquid, Motor vehicle (road), Mechanisms, Doors, and a variety of Domestic sounds. The full list of categories can be inspected in vocabulary.csv (see Files & Download below). The goal of the task was to build a multi-label audio tagging system that can predict appropriate label(s) for each audio clip in a test set.

What follows is a summary of some of the most relevant characteristics of FSDKaggle2019. Nevertheless, it is highly recommended to read our DCASE 2019 paper for a more in-depth description of the dataset and how it was built.

Ground Truth Labels

The ground truth labels are provided at the clip-level, and express the presence of a sound category in the audio clip, hence can be considered weak labels or tags. Audio clips have variable lengths (roughly from 0.3 to 30s).

The audio content from FSD has been manually labeled by humans following a data labeling process using the Freesound Annotator platform. Most labels have inter-annotator agreement but not all of them. More details about the data labeling process and the Freesound Annotator can be found in [2].

The YFCC soundtracks were labeled using automated heuristics applied to the audio content and metadata of the original Flickr clips. Hence, a substantial amount of label noise can be expected. The label noise can vary widely in amount and type depending on the category, including in- and out-of-vocabulary noises. More information about some of the types of label noise that can be encountered is available in [3].

Specifically, FSDKaggle2019 features three types of label quality, one for each set in the dataset:

curated train set: correct (but potentially incomplete) labels

noisy train set: noisy labels

test set: correct and complete labels

Further details can be found below in the sections for each set.

Format

All audio clips are provided as uncompressed PCM 16 bit, 44.1 kHz, mono audio files.

DATA SPLIT

FSDKaggle2019 consists of two train sets and one test set. The idea is to limit the supervision provided for training (i.e., the manually-labeled, hence reliable, data), thus promoting approaches to deal with label noise.

Curated train set

The curated train set consists of manually-labeled data from FSD.

Number of clips/class: 75 except in a few cases (where there are less)

Total number of clips: 4970

Avg number of labels/clip: 1.2

Total duration: 10.5 hours

The duration of the audio clips ranges from 0.3 to 30s due to the diversity of the sound categories and the preferences of Freesound users when recording/uploading sounds. Labels are correct but potentially incomplete. It can happen that a few of these audio clips present additional acoustic material beyond the provided ground truth label(s).

Noisy train set

The noisy train set is a larger set of noisy web audio data from Flickr videos taken from the YFCC dataset [5].

Number of clips/class: 300

Total number of clips: 19,815

Avg number of labels/clip: 1.2

Total duration: ~80 hours

The duration of the audio clips ranges from 1s to 15s, with the vast majority lasting 15s. Labels are automatically generated and purposefully noisy. No human validation is involved. The label noise can vary widely in amount and type depending on the category, including in- and out-of-vocabulary noises.

Considering the numbers above, the per-class data distribution available for training is, for most of the classes, 300 clips from the noisy train set and 75 clips from the curated train set. This means 80% noisy / 20% curated at the clip level, while at the duration level the proportion is more extreme considering the variable-length clips.

Test set

The test set is used for system evaluation and consists of manually-labeled data from FSD.

Number of clips/class: between 50 and 150

Total number of clips: 4481

Avg number of labels/clip: 1.4

Total duration: 12.9 hours

The acoustic material present in the test set clips is labeled exhaustively using the aforementioned vocabulary of 80 classes. Most labels have inter-annotator agreement but not all of them. Except human error, the label(s) are correct and complete considering the target vocabulary; nonetheless, a few clips could still present additional (unlabeled) acoustic content out of the vocabulary.

During the DCASE2019 Challenge Task 2, the test set was split into two subsets, for the public and private leaderboards, and only the data corresponding to the public leaderboard was provided. In this current package you will find the full test set with all the test labels. To allow comparison with previous work, the file test_post_competition.csv includes a flag to determine the corresponding leaderboard (public or private) for each test clip (see more info in Files & Download below).

Acoustic mismatch

As mentioned before, FSDKaggle2019 uses audio clips from two sources:

FSD: curated train set and test set, and

YFCC: noisy train set.

While the sources of audio (Freesound and Flickr) are collaboratively contributed and pretty diverse themselves, a certain acoustic mismatch can be expected between FSD and YFCC. We conjecture this mismatch comes from a variety of reasons. For example, through acoustic inspection of a small sample of both data sources, we find a higher percentage of high quality recordings in FSD. In addition, audio clips in Freesound are typically recorded with the purpose of capturing audio, which is not necessarily the case in YFCC.

This mismatch can have an impact in the evaluation, considering that most of the train data come from YFCC, while all test data are drawn from FSD. This constraint (i.e., noisy training data coming from a different web audio source than the test set) is sometimes a real-world condition.

LICENSE

All clips in FSDKaggle2019 are released under Creative Commons (CC) licenses. For attribution purposes and to facilitate attribution of these files to third parties, we include a mapping from the audio clips to their corresponding licenses.

Curated train set and test set. All clips in Freesound are released under different modalities of Creative Commons (CC) licenses, and each audio clip has its own license as defined by the audio clip uploader in Freesound, some of them requiring attribution to their original authors and some forbidding further commercial reuse. The licenses are specified in the files train_curated_post_competition.csv and test_post_competition.csv. These licenses can be CC0, CC-BY, CC-BY-NC and CC Sampling+.

Noisy train set. Similarly, the licenses of the soundtracks from Flickr used in FSDKaggle2019 are specified in the file train_noisy_post_competition.csv. These licenses can be CC-BY and CC BY-SA.

In addition, FSDKaggle2019 as a whole is the result of a curation process and it has an additional license. FSDKaggle2019 is released under CC-BY. This license is specified in the LICENSE-DATASET file downloaded with the FSDKaggle2019.doc zip file.

FILES & DOWNLOAD

FSDKaggle2019 can be downloaded as a series of zip files with the following directory structure:

root │
└───FSDKaggle2019.audio_train_curated/ Audio clips in the curated train set │ └───FSDKaggle2019.audio_train_noisy/ Audio clips in the noisy

REASSEMBLE: A Multimodal Dataset for Contact-rich Robotic Assembly and Disassembly

📋 Introduction

Robotic manipulation remains a core challenge in robotics, particularly for contact-rich tasks such as industrial assembly and disassembly. Existing datasets have significantly advanced learning in manipulation but are primarily focused on simpler tasks like object rearrangement, falling short of capturing the complexity and physical dynamics involved in assembly and disassembly. To bridge this gap, we present REASSEMBLE (Robotic assEmbly disASSEMBLy datasEt), a new dataset designed specifically for contact-rich manipulation tasks. Built around the NIST Assembly Task Board 1 benchmark, REASSEMBLE includes four actions (pick, insert, remove, and place) involving 17 objects. The dataset contains 4,551 demonstrations, of which 4,035 were successful, spanning a total of 781 minutes. Our dataset features multi-modal sensor data including event cameras, force-torque sensors, microphones, and multi-view RGB cameras. This diverse dataset supports research in areas such as learning contact-rich manipulation, task condition identification, action segmentation, and more. We believe REASSEMBLE will be a valuable resource for advancing robotic manipulation in complex, real-world scenarios.

{"Hama1": [
    [x ,y, z],
    [qx, qy, qz, qw]
 ], 
 "Hama2": [
    [x ,y, z],
    [qx, qy, qz, qw]
 ], 
 "DAVIS346": [
    [x ,y, z],
    [qx, qy, qz, qw]
 ], 
 "NIST_Board1": [
    [x ,y, z],
    [qx, qy, qz, qw]
 ]
}

[x, y, z] represent the position of the object, and [qx, qy, qz, qw] represent its orientation as a quaternion.

The HDF5 (.h5) format organizes data into two main types of structures: datasets, which hold the actual data, and groups, which act like folders that can contain datasets or other groups. In the diagram below, groups are shown as folder icons, and datasets as file icons. The main group of the file directly contains the video, audio, and event data. To save memory, video and audio are stored as encoded byte strings, while event data is stored as arrays. The robot’s proprioceptive information is kept in the robot_state group as arrays. Because different sensors record data at different rates, the arrays vary in length (signified by the N_xxx variable in the data shapes). To align the sensory data, each sensor’s timestamps are stored separately in the timestamps group. Information about action segments is stored in the segments_info group. Each segment is saved as a subgroup, named according to its order in the demonstration, and includes a start timestamp, end timestamp, a success indicator, and a natural language description of the action. Within each segment, low-level skills are organized under a low_level subgroup, following the same structure as the high-level annotations.

📁

The splits folder contains two text files which list the h5 files used for the traning and validation splits.

📌 Important Resources

The project website contains more details about the REASSEMBLE dataset. The Code for loading and visualizing the data is avaibile on our github repository.

📄 Project website: https://tuwien-asl.github.io/REASSEMBLE_page/
💻 Code: https://github.com/TUWIEN-ASL/REASSEMBLE

MusicNet is a collection of 330 freely-licensed classical music recordings, together with over 1 million annotated labels indicating the precise time of each note in every recording, the instrument that plays each note, and the note's position in the metrical structure of the composition. The labels are acquired from musical scores aligned to recordings by dynamic time warping. The labels are verified by trained musicians; we estimate a labeling error rate of 4%. We offer the MusicNet labels to the machine learning and music communities as a resource for training models and a common benchmark for comparing results. This dataset was introduced in the paper "Learning Features of Music from Scratch." [1]

This repository consists of 3 top-level files:

• musicnet.tar.gz - This file contains the MusicNet dataset itself, consisting of PCM-encoded audio wave files (.wav) and corresponding CSV-encoded note label files (.csv). The data is organized according to the train/test split described and used in "Invariances and Data Augmentation for Supervised Music Transcription". [2]
• musicnet_metadata.csv - This file contains track-level information about recordings contained in MusicNet. The data and label files are named with MusicNet ids, which you can use to cross-index the data and labels with this metadata file.
• musicnet_midis.tar.gz - This file contains the reference MIDI files used to construct the MusicNet labels.

A PyTorch interface for accessing the MusicNet dataset is available on GitHub. For an audio/visual introduction and summary of this dataset, see the MusicNet inspector, created by Jong Wook Kim. The audio recordings in MusicNet consist of Creative Commons licensed and Public Domain performances, sourced from the Isabella Stewart Gardner Museum, the European Archive Foundation, and Musopen. The provenance of specific recordings and midis are described in the metadata file.

[1] Learning Features of Music from Scratch. John Thickstun, Zaid Harchaoui, and Sham M. Kakade. In International Conference on Learning Representations (ICLR), 2017. ArXiv Report.

@inproceedings{thickstun2017learning,
  title={Learning Features of Music from Scratch},
  author = {John Thickstun and Zaid Harchaoui and Sham M. Kakade},
  year={2017},
  booktitle = {International Conference on Learning Representations (ICLR)}
}

[2] Invariances and Data Augmentation for Supervised Music Transcription. John Thickstun, Zaid Harchaoui, Dean P. Foster, and Sham M. Kakade. In International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2018. ArXiv Report.

@inproceedings{thickstun2018invariances,
title={Invariances and Data Augmentation for Supervised Music Transcription},
author = {John Thickstun and Zaid Harchaoui and Dean P. Foster and Sham M. Kakade},
year={2018},
booktitle = {International Conference on Acoustics, Speech, and Signal Processing (ICASSP)}
}

Context

As my final project in Data Analytics Bootcamp I decided to build cocktail generator. Dataset contains of almost 500 alcoholic cocktails where each ingredient has asigned label in column 'Alc-type' or 'Basic-taste' and defined volume in ml or grams (columns 'Value-ml' & 'Value-gr').

Content

Dataset prepared by AIFirst has been used as a basis for this dataset: Cocktails Ingredients

Inspiration

I hope this cleaned and organized dataset will become useful for analysis or modeling.

Confusion matrix for the randomized label set used to control for the performance of TDANet and ResNet. The (i,j)th entry represents the number of colonies whose actual label was i, but were assigned label j. *There was one BMP4 colony randomly labeled as DS+CHIR, for which we had colony images but lacked the cell coordinate data. Thus, while we could use this colony for training ResNet, it was not available for use by TDANet.

AIT Log Data Sets

This repository contains synthetic log data suitable for evaluation of intrusion detection systems. The logs were collected from four independent testbeds that were built at the Austrian Institute of Technology (AIT) following the approach by Landauer et al. (2020) [1]. Please refer to the paper for more detailed information on automatic testbed generation and cite it if the data is used for academic publications. In brief, each testbed simulates user accesses to a webserver that runs Horde Webmail and OkayCMS. The duration of the simulation is six days. On the fifth day (2020-03-04) two attacks are launched against each web server.

The archive AIT-LDS-v1_0.zip contains the directories "data" and "labels".

The data directory is structured as follows. Each directory mail.

Setup details of the web servers:

• OS: Debian Stretch 9.11.6
• Services:
  • Apache2
  • PHP7
  • Exim 4.89
  • Horde 5.2.22
  • OkayCMS 2.3.4
  • Suricata
  • ClamAV
  • MariaDB

Setup details of user machines:

• OS: Ubuntu Bionic
• Services:
  • Chromium
  • Firefox

User host machines are assigned to web servers in the following way:

• mail.cup.com is accessed by users from host machines user-{0, 1, 2, 6}
• mail.spiral.com is accessed by users from host machines user-{3, 5, 8}
• mail.insect.com is accessed by users from host machines user-{4, 9}
• mail.onion.com is accessed by users from host machines user-{7, 10}

The following attacks are launched against the web servers (different starting times for each web server, please check the labels for exact attack times):

• Attack 1: multi-step attack with sequential execution of the following attacks:
  • nmap scan
  • nikto scan
  • smtp-user-enum tool for account enumeration
  • hydra brute force login
  • webshell upload through Horde exploit (CVE-2019-9858)
  • privilege escalation through Exim exploit (CVE-2019-10149)
• Attack 2: webshell injection through malicious cookie (CVE-2019-16885)

Attacks are launched from the following user host machines. In each of the corresponding directories user-

• user-6 attacks mail.cup.com
• user-5 attacks mail.spiral.com
• user-4 attacks mail.insect.com
• user-7 attacks mail.onion.com

The log data collected from the web servers includes

• Apache access and error logs
• syscall logs collected with the Linux audit daemon
• suricata logs
• exim logs
• auth logs
• daemon logs
• mail logs
• syslogs
• user logs

Note that due to their large size, the audit/audit.log files of each server were compressed in a .zip-archive. In case that these logs are needed for analysis, they must first be unzipped.

Labels are organized in the same directory structure as logs. Each file contains two labels for each log line separated by a comma, the first one based on the occurrence time, the second one based on similarity and ordering. Note that this does not guarantee correct labeling for all lines and that no manual corrections were conducted.

Version history and related data sets:

• AIT-LDS-v1.0: Four datasets, logs from single host, fine-granular audit logs, mail/CMS.
  • AIT-LDS-v1.1: Removed carriage return of line endings in audit.log files.
• AIT-LDS-v2.0: Eight datasets, logs from all hosts, system logs and network traffic, mail/CMS/cloud/web.

Acknowledgements: Partially funded by the FFG projects INDICAETING (868306) and DECEPT (873980), and the EU project GUARD (833456).

If you use the dataset, please cite the following publication:

[1] M. Landauer, F. Skopik, M. Wurzenberger, W. Hotwagner and A. Rauber, "Have it Your Way: Generating Customized Log Datasets With a Model-Driven Simulation Testbed," in IEEE Transactions on Reliability, vol. 70, no. 1, pp. 402-415, March 2021, doi: 10.1109/TR.2020.3031317. [PDF]

Dataset description

This dataset consists of 74 patients age 4-51 years old where Cortico-Cortical Evoked Potentials (CCEPs) were measured with Electro-CorticoGraphy (ECoG) during single pulse electrical stimulation. For a detailed description see:

• Developmental trajectory of transmission speed in the human brain. D. van Blooijs¹, M.A. van den Boom¹, J.F. van der Aar, G.J.M. Huiskamp, G. Castegnaro, M. Demuru, W.J.E.M. Zweiphenning, P. van Eijsden, K. J. Miller, F.S.S. Leijten, D. Hermes, Nature Neuroscience, 2023, https://doi.org/10.1038/s41593-023-01272-0
  ¹ these authors contributed equally.

This dataset is part of the RESPect (Registry for Epilepsy Surgery Patients) database, a dataset recorded at the University Medical Center of Utrecht, the Netherlands. The study was approved by the Medical Ethical Committee from the UMC Utrecht.

Contact

• Dorien van Blooijs: D.vanBlooijs@umcutrecht.nl
• Frans Leijten: F.S.S.leijten@umcutrecht.nl
• Dora Hermes: hermes.dora@mayo.edu

Data organization

This data is organized according to the Brain Imaging Data Structure specification. A community-driven specification for organizing neurophysiology data along with its metadata. For more information on this data specification, see https://bids-specification.readthedocs.io/en/stable/

Each patient has their own folder (e.g., sub-ccepAgeUMCU01 to sub-ccepAgeUMCU74) which contains the iEEG recordings data for that patient, as well as the metadata needed to understand the raw data and event timing.

Data are logically grouped in the same BIDS session and stored across runs that indicating the day and time point of recording during the monitoring period. If extra electrodes were added/removed during this period, the session was divided into different sessions (e.g. ses-1a and ses-1b). We use the optional run key-value pair to specify the day and the start time of the recording (e.g. run-021315, day 2 after implantation, which is day 1 of the monitoring period, at 13:15). The task key-value pair in long-term iEEG recordings describes the patient's state during the recording of this file. The task label is “SPESclin“ since these files contain data collected during clinical single pulse electrical stimulation (SPES).

Electrode positions include Destrieux atlas labels that were estimated by running Freesurfer on the individual subject MRI scan and taking the most common surface label within a sphere around the electrode. All shared electrode positions were then converted to MNI152 space using the Freesurfer surface based non-linear transformation. We note that this surface based transformation distorts the dimensions of the grids, but maintains the gyral anatomy.

License

This dataset is made available under the Public Domain Dedication and License CC v1.0, whose full text can be found at https://creativecommons.org/publicdomain/zero/1.0/. We hope that all users will follow the ODC Attribution/Share-Alike Community Norms (http://www.opendatacommons.org/norms/odc-by-sa/); in particular, while not legally required, we hope that all users of the data will acknowledge by citing the following in any publication: Developmental trajectory of transmission speed in the human brain, D. van Blooijs, M.A. van den Boom, J.F. van der Aar, G.J.M. Huiskamp, G. Castegnaro, M. Demuru, W.J.E.M. Zweiphenning, P. van Eijsden, K. J. Miller, F.S.S. Leijten, D. Hermes, Nature Neuroscience, 2023, https://doi.org/10.1038/s41593-023-01272-0

Code

Code to analyses these data is available at: https://github.com/MultimodalNeuroimagingLab/mnl_ccepAge

Acknowledgements

We thank the SEIN-UMCU RESPect database group (C.J.J. van Asch, L. van de Berg, S. Blok, M.D. Bourez, K.P.J. Braun, J.W. Dankbaar, C.H. Ferrier, T.A. Gebbink, P.H. Gosselaar, R. van Griethuysen, M.G.G. Hobbelink, F.W.A. Hoefnagels, N.E.C. van Klink, M.A. van ‘t Klooster, G.A.P. deKort, M.H.M. Mantione, A. Muhlebner, J.M. Ophorst, P.C. van Rijen, S.M.A. van der Salm, E.V. Schaft, M.M.J. van Schooneveld, H. Smeding, D. Sun, A. Velders, M.J.E. van Zandvoort, G.J.M. Zijlmans, E. Zuidhoek and J. Zwemmer) for their contributions and help in collecting the data, and G. Ojeda Valencia for proofreading the manuscript.

Funding

Research reported in this publication was supported by the National Institute of Mental Health of the National Institutes of Health under Award Number R01MH122258 (DH, FSSL, the content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health), the EpilepsieNL under Award Number NEF17-07 (DvB) and the UMC Utrecht Alexandre Suerman MD/PhD Stipendium 2015 (WZ).

The Mathematics Subject Classification organizes Publications, Software, and Research Data into a hierarchical classification scheme maintained by MathSciNet (mr) and zbMATH Open (zbmath). According to the classification scheme, both organizations mr and zbmath agree on this classification and use labels to organize publications from mathematics and related fields. However, the classification of individual papers is done independently of each other. This dataset contains references to papers that occur in both collections (mr and zbmath) together with the respective classification labels.

The dataset is followed in the follwing form

zbmath-id, zbmath-msc, mr-id, mr-msc
5635019, 55-06 57-06 55R70 57Q45 00B25, MR2556072, 54-06 55-06
5641347, 68R10 05C85, MR2588354, 68W25 05C70 05C85
5641348, 68R10, MR2588355, 68Q25 05C65 05C70 05C85 68Q15
5641349, 68Q05, MR2588356, 68Q05
5641350, 68M20 68W25 68T42, MR2588357, 68T42 68W25
5641351, 68M10, MR2588358, 68Q85 68Q10
5641352, 68R15, MR2588359, 68R15 05A05 05C78
5641353, 68T30 68R10, MR2588360, 05C62 68R10
5641354, 68Q30, MR2588361, 68Q30 60A99 60J20
5641355, 68W27 68M10, MR2588362, 68M10 05C82 05C85 68W27 68W40
5641356, 68W05 68T05, MR2588363, 68T05 62H30
5641357, 68W40, MR2588364, 05A15 68R05
5641358, 91A10 91A05 68Q17 91A06, MR2588365, 91A05 68W25
5641359, 91A10 68T42 68M10, MR2588366, 91B26
5641360, 68W40 68P05 68P10, MR2588367, 68W40 68P05 68Q87
5641361, 68P25 94A62, MR2588368, 94A62 11T71 68P25
5641362, 68Q45, MR2588369, 68Q45 05A05
5641363, 90C35, MR2588370, 68R10 05C85 68W25 90C35
5641364, 54H20, MR2588371, 37E10 37B10 37E45

The meaning of the fields is:

• zbmath-id Unique identifier from zbMATH Open. Prefix with https://zbmath.org/ to visit additional information on the article. For example, 5635019 is associated with https://zbmath.org/5635019
• zbmath-msc space separated list of Mathemematics Subject Classifiction labels created by zbMATH Open staff. A description of the label can be retrieved by prefixing https://zbmath.org/classification/?q=. For example, 55-06 is associated with https://zbmath.org/classification/?q=55-06
• mr-id Unique identifier from MathReviews. Prefix with https://mathscinet.ams.org/mathscinet-getitem?mr= to retrieve additional information on the publication. For example, MR3844132 is associated with https://mathscinet.ams.org/mathscinet-getitem?mr=MR2556072.
• mr-msc space seperated list of Mathemematics Subject Classifiction labels created by MathSciNet staff.

The dataset was retrieved in 2016 by querying MathSciNet and zbMATH Open. Therefore, the classifications are based on the MSC 2010 version.

This dataset was used in

Schubotz M., Scharpf P., Teschke O., Kühnemund A., Breitinger C., Gipp B. (2020) AutoMSC: Automatic Assignment of Mathematics Subject Classification Labels. In: Benzmüller C., Miller B. (eds) Intelligent Computer Mathematics. CICM 2020. Lecture Notes in Computer Science, vol 12236. Springer, Cham. https://doi.org/10.1007/978-3-030-53518-6_15

and is now released to the public as an addendum to the paper.

Dataset Description for Chest X-rays (Montgomery and Shenzhen)

https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F7162438%2F8ca0d9e4c8d1853a14b9e20e6f600b97%2F_results_11_2.png?generation=1732001019088722&alt=media" alt=""> This dataset consists of 704 chest X-ray images that have been curated from two sources: the Montgomery County Chest X-ray Database (USA) and the Shenzhen Chest X-ray Database (China). The images are used for training and evaluating machine learning models for tuberculosis (TB) detection.

The dataset contains both tuberculosis-positive and normal chest X-rays, along with demographic details such as gender, age, and county of origin. The images are accompanied by lung segmentation masks and clinical metadata, which makes the dataset highly suitable for deep learning applications in medical imaging.

Dataset Overview

Data Sources

• Montgomery County Chest X-ray Database
• Shenzhen Chest X-ray Database

Dataset Stats

1. Total Images: 704 chest X-rays (from both Montgomery and Shenzhen).
2. County Distribution: - Shenzhen: 80% (563 images). - Montgomery: 20% (141 images).

3. PTB (Tuberculosis Cases): - PTB=1 (Tuberculosis Positive): 345 images. - PTB=0 (Normal): 359 images.

Clinical Data Breakdown:

The clinical metadata file includes the following columns: - id: Unique identifier for each image. - gender: Gender of the patient. - age: Age of the patient. - county: County of origin (Shenzhen or Montgomery). - ptb: Label indicating if the image shows tuberculosis (PTB=1) or is normal (PTB=0). - remarks: Additional clinical notes about the patient’s condition (e.g., "secondary PTB", "normal").

Organized Data Structure:

/datasets/
  /image/      # X-ray images
  /mask/      # Lung segmentation masks
  /MetaData.csv   # Clinical metadata

Next Steps and Recommendations

To improve the dataset and address some of the challenges: - Balance the dataset by using oversampling/undersampling techniques or generate synthetic data for underrepresented categories. - Increase data diversity: Consider adding more datasets from different regions and with different demographic distributions. - Use transfer learning for model training, leveraging models pretrained on larger datasets (e.g., ImageNet) to overcome the small dataset size.

With fast growth, synthetic biology powers us with the capability to produce high commercial value products in an efficient resource/energy-consuming manner. Comprehensive knowledge of the protein regulatory network of a bacterial host chassis, e.g., the actual amount of the given proteins, is the key to building cell factories for certain target hyperproduction. Many talent methods have been introduced for absolute quantitative proteomics. However, for most cases, a set of reference peptides with isotopic labeling (e.g., SIL, AQUA, QconCAT) or a set of reference proteins (e.g., commercial UPS2 kit) needs to be prepared. The higher cost hinders these methods for large sample research. In this work, we proposed a novel metabolic labeling-based absolute quantification approach (termed nMAQ). The reference Corynebacterium glutamicum strain is metabolically labeled with 15N, and a set of endogenous anchor proteins of the reference proteome is quantified by chemically synthesized light (14N) peptides. The prequantified reference proteome was then utilized as an internal standard (IS) and spiked into the target (14N) samples. SWATH-MS analysis is performed to obtain the absolute expression levels of the proteins from the target cells. The cost for nMAQ is estimated to be less than 10 dollars per sample. We have benchmarked the quantitative performance of the novel method. We believe this method will help with the deep understanding of the intrinsic regulatory mechanism of C. glutamicum during bioengineering and will promote the process of building cell factories for synthetic biology.

Image Tagging and Annotation Services Market Outlook

The global image tagging and annotation services market size was valued at approximately USD 1.5 billion in 2023 and is projected to reach around USD 4.8 billion by 2032, growing at a compound annual growth rate (CAGR) of about 14%. This robust growth is driven by the exponential rise in demand for machine learning and artificial intelligence applications, which heavily rely on annotated datasets to train algorithms effectively. The surge in digital content creation and the increasing need for organized data for analytical purposes are also significant contributors to the market expansion.

One of the primary growth factors for the image tagging and annotation services market is the increasing adoption of AI and machine learning technologies across various industries. These technologies require large volumes of accurately labeled data to function optimally, making image tagging and annotation services crucial. Specifically, sectors such as healthcare, automotive, and retail are investing in AI-driven solutions that necessitate high-quality annotated images to enhance machine learning models' efficiency. For example, in healthcare, annotated medical images are essential for developing tools that can aid in diagnostics and treatment decisions. Similarly, in the automotive industry, annotated images are pivotal for the development of autonomous vehicles.

Another significant driver is the growing emphasis on improving customer experience through personalized solutions. Companies are leveraging image tagging and annotation services to better understand consumer behavior and preferences by analyzing visual content. In retail, for instance, businesses analyze customer-generated images to tailor marketing strategies and improve product offerings. Additionally, the integration of augmented reality (AR) and virtual reality (VR) in various applications has escalated the need for precise image tagging and annotation, as these technologies rely on accurately labeled datasets to deliver immersive experiences.

Data Collection and Labeling are foundational components in the realm of image tagging and annotation services. The process of collecting and labeling data involves gathering vast amounts of raw data and meticulously annotating it to create structured datasets. These datasets are crucial for training machine learning models, enabling them to recognize patterns and make informed decisions. The accuracy of data labeling directly impacts the performance of AI systems, making it a critical step in the development of reliable AI applications. As industries increasingly rely on AI-driven solutions, the demand for high-quality data collection and labeling services continues to rise, underscoring their importance in the broader market landscape.

The rising trend of digital transformation across industries has also significantly bolstered the demand for image tagging and annotation services. Organizations are increasingly investing in digital tools that can automate processes and enhance productivity. Image annotation plays a critical role in enabling technologies such as computer vision, which is instrumental in automating tasks ranging from quality control to inventory management. Moreover, the proliferation of smart devices and the Internet of Things (IoT) has led to an unprecedented amount of image data generation, further fueling the need for efficient image tagging and annotation services to make sense of the vast data deluge.

From a regional perspective, North America is currently the largest market for image tagging and annotation services, attributed to the early adoption of advanced technologies and the presence of numerous tech giants investing in AI and machine learning. The region is expected to maintain its dominance due to ongoing technological advancements and the growing demand for AI solutions across various sectors. Meanwhile, the Asia Pacific region is anticipated to experience the fastest growth during the forecast period, driven by rapid industrialization, increasing internet penetration, and the rising adoption of AI technologies in countries like China, India, and Japan. The European market is also witnessing steady growth, supported by government initiatives promoting digital innovation and the use of AI-driven applications.

Service Type Analysis

The service type segment in the image tagging and annotation services market is bifurcated into manual annotation and automa

The results presented in Fig 6 are averaged over all percentages of labeled data. Best performance for each dataset is in bold.

Plasmids play an essential role in horizontal gene transfer, aiding their host bacteria in acquiring beneficial traits like antibiotic and metal resistance. There exists some plasmids that can transfer, replicate or persist in multiple organisms. Identifying the relatively complete host range of these plasmids provides insights into how plasmids promote bacterial evolution. To achieve this, we can apply multi-label learning models for plasmid host range prediction. However, there are no databases providing the detailed and complete host labels of these broad-host-range (BHR) plasmids. Without adequate well-annotated training samples, learning models can fail to extract discriminative feature representations for plasmid host prediction.

To address this problem, we propose a self-correction multi-label learning model called MOSTPLAS. We design a pseudo label learning algorithm and a self-correction asymmetric loss to facilitate the training of multi-label learning model with samples containing some unknown missing labels. We conducted a series of experiments on NCBI RefSeq plasmid database, plasmids with experimentally determined host labels, Hi-C dataset and DoriC dataset. The benchmark results against other plasmid host range prediction tools demonstrated that MOSTPLAS recognized more host labels while keeping a high precision.

This repository holds the data and models created by training and testing a Hybrid Deep Learning model whose results are published in the Conference Paper "To Whom are You Talking? A DL model to Endow Social Robots with Addressee Estimation Skills" presented at the International Joint Conference on Neural Networks (IJCNN) 2023. https://ieeexplore.ieee.org/document/10191452
OA version: http://doi.org/10.48550/arXiv.2308.10757/

Addressee Estimation is the ability to understand to whom a person is directing an utterance. This ability is crucial for social robot engaging in multi-party interaction to understand the basic dynamics of social communication.
In this project, we trained a DL model composed of convolutional layers and LSTM cells and taking as input visual information of the speaker to estimate the placement of the addressee. We used a supervised learning approach. The data to train our model were taken from the Vernissage Corpus, a dataset collected in multi-party Human-Robot Interaction from the robot's sensors. For the original HRI corpus, see http://vernissage.humavips.eu/

This repository contains the /labels used for the training, the /models resulting from the training, and the /resuls, i.e., the files containing the results of tests.

The codes to obtain this data can be found at http://doi.org/10.5281/zenodo.10709858

• The folder /labels contains the .csv files that are used for the supervised training. The sheets list the frames extracted from the videos of the Vernissage Corpus temporally organized in sequences of 10 frames (the minimum chunk of data the hybrid CNN-LSTM model is trained on).
  Labels are divided in two folders: 3CLASSES (to train models classifying the addressee in terms of left/right/nao) or BINARY (to train models with a binary classification (the robot is involved in the conversation or not). The Vernissage corpus was annotated with four possible Addressees: left/right/nao/group. In this work we used the first three for the 3-class classification and we grouped "left" and "right" together into "not-involved" and "nao" and "grouped" together into "involved" for the binary classification. (see paper for more details).
  The sheets connect each frame with:
  • their ground truth:
    • LABEL (int)
    • ADDRESSEE (string)
  • the number of the SEQUENCE (int)
  • the number of the SLOT of the Vernissage Dataset the frame is extracted from (int)
  • the number of the SPEAKER, because for each slot of the Vernissage Corpus there are two speakers (int)
  • the number of the interval (N_INTERVAL), representing the number of the utterance, i.e. the interval of speech divided by at least 0.8 s of silence (each interval/utterance can contains one or more sequences)
  • temporal information to capture the frame from the video of the slot: 4 temporal information are given
    • START_MSEC: start of the capture (for audio -> not used in this project)
    • END_MSEC: stop of the capture (for audio -> not used in this project)
    • FRAME_MSEC: exact time of the capture (for image)
    • DURATION_MSEC: the duration in msec of the frame is referred to (sequences are made of 10 frames, each captured every 80 msec)
  • the filename of the frame (the image captured from the video): IMG (.jpg)
  • the filename of the speaker's face image cropped from the frame: FACE_IMAGE (.jpg)
  • the filename of the speaker's pose vector extracted from the frame: FACE_POSE (.npy)
  • the filename of the face image of the second person cropped from the frame: FACE_OTHER (.jpg) (Not used)
  • the filename of the pose vector of the second person extracted from the frame: FACE_OTHER (.npy) (Not used)
  • the filename of the audio captured from START_MSEC to END_MSEC (Not used)
  • the filename of the audio captured from START_MSEC to END_MSEC (Not used)
  • the filename of the mel spectrogram extracted from the audio (Not used)
  • the filename of the MFCC features extracted from the audio (Not used)
    
    TIP: since label files contains temporal information about when frames, faces and pose were captured, cropped and extracted from the original naovideo.avi contained in each slot of the vernissage corpus, in order to obtain the dataset of this project from the vernissage corpus, it is possible to use these label files, capture frames with the temporal information provided in the .csv file and then crop faces and extract poses as explained in the code connected to this dataset.

• The folder /models contains three subfolders: /three_class , /binary and /complete
  In the /three_class folder there are models obtained from the 10-fold cross-validation explained in the paper, using 4 different architectures:
  • hybrid CNN-LSTM taking in input the speaker's face and pose with intermediate fusion before the LSTM
  • hybrid CNN-LSTM taking in input the speaker's face and pose with late fusion after the LSTM
  • hybrid CNN-LSTM taking in input only the speaker's face
  • hybrid CNN-LSTM taking in input only the speaker's pose
    
    Each of them contains 10 folders (coming from the 10 fold cross validation) with the .pth files obtained from the training and the plots graphically representing the training Loss and Accuracy.
    
    The /binary folder containes the same files but only for the hybrid CNN-LSTM taking in input the speaker's face and pose with intermediate fusion before the LSTM
    
    The /complete folder contains the hybrid CNN-LSTM taking in input the speaker's face and pose with intermediate fusion before the LSTM trained on the entire vernissage corpus

• The folder /results is divided in /three_class and /binary as well.
  /three_class is divided in the same subfolder as /models. Each of them contains .csv files of the results coming from the testing phase on the only slot left out of the training.
  • In results.csv, for every sequence of the slot left out for the test, is given also the ground truth (label), the prediction of the model after training (predictions) and the score/confidence of the prediction (score)
  • In errors.csv are listed the sequence that have been classified badly by the model. label and prediction are given for each of them.
  • the .npy file provides information about the model

• The folder /dataset_slots contains 10 empty folders, you can use those to organize the Vernissage Dataset, according to the code released in http://doi.org/10.5281/zenodo.10709858

Data Description

This dataset is linked to the publication "Recursive classification of satellite imaging time-series: An application to land cover mapping". In this paper, we introduce the recursive Bayesian classifier (RBC), which converts any instantaneous classifier into a robust online method through a probabilistic framework that is resilient to non-informative image variations. To reproduce the results presented in the paper, the RBC-SatImg folder and the code in the GitHub repository RBC-SatImg are required.

The RBC-SatImg folder contains:

• Sentinel-2 time-series imagery from three key regions: Oroville Dam (CA, USA) and Charles River (Boston, MA, USA) for water mapping, and the Amazon Rainforest (Brazil) for deforestation detection.
• The RBC-WatData dataset with manually generated water mapping labels for the Oroville Dam and Charles River regions. This dataset is well-suited for multitemporal land cover and water mapping research, as it accounts for the dynamic evolution of true class labels over time.
• Pickle files with output to reproduce the results in the paper, including:
  • Instantaneous classification results for GMM, LR, SIC, WN, DWM
  • Posterior results obtained with the RBC framework

The Sentinel-2 images and forest labels used in the deforestation detection experiment for the Amazon Rainforest have been obtained from the MultiEarth Challenge dataset.

Folder Structure

The following paths can be changed in the configuration file from the GitHub repository as desired. The RBC-SatImg is organized as follows:

• `./log/` (EMPTY): Default path for storing log files generated during code execution.
• `./evaluation_results/`: Contains the results to reproduce the findings in the paper, including two sub-folders:
  • `./classification/`: For each test site, four sub-folders are included as:
    • `./accuracy/`: Each sub-folder corresponding to an experimental configuration contains pickle files with balanced classification accuracy results and information about the models. The default configuration used in the paper is "conf_00."
    • `./figures/`: Includes result figures from the manuscript in SVG format.
    • `./likelihoods/`: Contains pickle files with instantaneous classification results.
    • `./posteriors/`: Contains pickle files with posterior results generated by the RBC framework.
  • `./sensitivity_analysis/`: Contains sensitivity analysis results, organized by different test sites and epsilon values.
• `./Sentinel2_data/`: Contains Sentinel-2 images used for training and evaluation, organized by scenarios (Oroville Dam, Charles River, Amazon Rainforest). Selected images have been filtered and processed as explained in the manuscript. The Amazon Rainforest images and labels have been obtained from the MultiEarth dataset, and consequently, the labels are included in this folder instead of the RBC-WatData folder.
• `./RBC-WatData/`: Contains the water labels that we manually generated with the LabelStudio tool.

Description of the different characteristics of benchmark datasets: Size, number of features, and number of classes.