Open Dataset Research Articles

Sensitive data identification for multi‐category and multi‐scenario data

AbstractSensitive data identification is the prerequisite for protecting critical user and business data. Traditional methods usually only target a certain type of application scenario or a certain type of data, thus making it difficult to meet the needs of enterprise‐level data protection. This paper proposes an introduction to the end‐to‐end sensitive data identification system of Beike Inc. The system consists of the data identification & annotation platform, dataset management platform, and sensitive data identification model, which propose different governance methods for batch data and streaming data respectively. Specifically, we propose a sliding window‐based identification method for long text to improve the identification of streaming data. Evaluation results show that this method can improve the effect of identifying long text sensitive data without losing the ability on short text, for the open source test dataset, the value can be up to 94.15, so it is applicable in diverse scenarios.

Comparison of deep learning models for milk production forecasting at national scale

Forecasting of agricultural variables at national scale is essential for developing policies aimed at guaranteeing food security and increase the stability of the entire agri-food chain. However, this is often a challenging task due to the fact that data are usually sparse and long records are rarely available. This work aims at evaluating the potentiality of ten deep learning models for predicting monthly milk production in France, Germany, and Italy, using as input, climatic and economic variables from open datasets. The results indicate that deep learning models are a better alternative to traditional statistical approaches, providing robust results without requiring complex model architectures. Furthermore, the prominent autoregressive nature of milk production highlights the inability of environmental variables to capture external processes that influence the milk production (e.g., milk quota). However, the high accuracy achieved by the models lays the ground for possible application of deep learning models to accurately forecast agricultural variables at national scale with potential implications for the development of dairy insurance products, and risk management practices.

Blind Federated Learning without initial model

Federated learning is an emerging machine learning approach that allows the construction of a model between several participants who hold their own private data. This method is secure and privacy-preserving, suitable for training a machine learning model using sensitive data from different sources, such as hospitals. In this paper, the authors propose two innovative methodologies for Particle Swarm Optimisation-based federated learning of Fuzzy Cognitive Maps in a privacy-preserving way. In addition, one relevant contribution this research includes is the lack of an initial model in the federated learning process, making it effectively blind. This proposal is tested with several open datasets, improving both accuracy and precision.

RegWSI: Whole slide image registration using combined deep feature- and intensity-based methods: Winner of the ACROBAT 2023 challenge

Background and objectiveThe automatic registration of differently stained whole slide images (WSIs) is crucial for improving diagnosis and prognosis by fusing complementary information emerging from different visible structures. It is also useful to quickly transfer annotations between consecutive or restained slides, thus significantly reducing the annotation time and associated costs. Nevertheless, the slide preparation is different for each stain and the tissue undergoes complex and large deformations. Therefore, a robust, efficient, and accurate registration method is highly desired by the scientific community and hospitals specializing in digital pathology. MethodsWe propose a two-step hybrid method consisting of (i) deep learning- and feature-based initial alignment algorithm, and (ii) intensity-based nonrigid registration using the instance optimization. The proposed method does not require any fine-tuning to a particular dataset and can be used directly for any desired tissue type and stain. The registration time is low, allowing one to perform efficient registration even for large datasets. The method was proposed for the ACROBAT 2023 challenge organized during the MICCAI 2023 conference and scored 1st place. The method is released as open-source software. ResultsThe proposed method is evaluated using three open datasets: (i) Automatic Nonrigid Histological Image Registration Dataset (ANHIR), (ii) Automatic Registration of Breast Cancer Tissue Dataset (ACROBAT), and (iii) Hybrid Restained and Consecutive Histological Serial Sections Dataset (HyReCo). The target registration error (TRE) is used as the evaluation metric. We compare the proposed algorithm to other state-of-the-art solutions, showing considerable improvement. Additionally, we perform several ablation studies concerning the resolution used for registration and the initial alignment robustness and stability. The method achieves the most accurate results for the ACROBAT dataset, the cell-level registration accuracy for the restained slides from the HyReCo dataset, and is among the best methods evaluated on the ANHIR dataset. ConclusionsThe article presents an automatic and robust registration method that outperforms other state-of-the-art solutions. The method does not require any fine-tuning to a particular dataset and can be used out-of-the-box for numerous types of microscopic images. The method is incorporated into the DeeperHistReg framework, allowing others to directly use it to register, transform, and save the WSIs at any desired pyramid level (resolution up to 220k x 220k). We provide free access to the software. The results are fully and easily reproducible. The proposed method is a significant contribution to improving the WSI registration quality, thus advancing the field of digital pathology.

3D medical image segmentation based on semi-supervised learning using deep co-training

In recent years, artificial intelligence has been applied to 3D COVID-19 medical image diagnosis, which reduces detection costs and missed diagnosis rates with higher predictive accuracy, and diagnostic efficiency. However, the limited size and low quality of clinical 3D medical image samples have hindered the segmentation performance of 3D models. Therefore, we propose a 3D medical image segmentation model based on semi-supervised learning using co-training. Multi-view and multi-modal images are generated using spatial flipping and windowing techniques to enhance the spatial diversity of 3D image samples. A pseudo label generation module based on confidence-weights is employed to generate reliable pseudo labels for non-annotated data, thereby increasing the sample size and reducing overfitting. The proposed approach utilizes a three-stage training process: firstly, training a single network based on annotated data; secondly, incorporating non-annotated data to train a dual-modal network and generate pseudo labels; finally, jointly training six models in three dimensions using both annotated and pseudo labels generated from multi-view and multi-modal images, aiming to enhance segmentation accuracy and generalization performance. Additionally, a consistency regularization loss is applied to reduce noises and accelerate convergence of the training. Moreover, a heatmap visualization method is employed to focus on the attention of features at each stage of training, providing effective reference for clinical diagnosis. Experiments were conducted on an open dataset of 3D COVID-19 CT samples and a non-annotated dataset from TCIA, including 771 NIFTI-format CT images from 661 COVID-19 patients. The results of 5-fold cross-validation show that the proposed model achieves a segmentation accuracy of Dice=73.30 %, ASD=10.633, Sensitivity=63.00 %, and Specificity=99.60 %. Compared to various typical semi-supervised learning 3D segmentation models, it demonstrates better segmentation accuracy and generalization performance.

Detection of Wheeze Sounds in Respiratory Disorders: A Deep Learning Approach

Respiratory disorders, including chronic obstructive pulmonary disease (COPD) and asthma, are major causes of death globally. Early diagnosis of these conditions is essential for effective treatment. Auscultation of the lungs is the traditional diagnostic method, which has drawbacks such as subjectivity and susceptibility to environmental interference. To overcome these limitations, this study presents a novel approach for wheeze detection using deep learning methods. This approach includes the usage of artificial data created by employing the open ICBHI dataset with the aim of improvement in generalization of learning models. Spectrograms that were obtained as the output of the Short-Time Fourier Transform analysis were employed in feature extraction. Two labeling approaches were used for model comparison. The first approach involved labeling after wheezing occurred, and the second approach assigned labels directly to the time steps where wheezing patterns are seen. Wheeze event detection was performed by constructing four RNN-based models (CNN-LSTM, CNN-GRU, CNN-BiLSTM, and CNN-BiGRU). It was observed that labeling wheeze events directly resulted in more precise detection, with exceptional performance exhibited by the CNN-BiLSTM model. This approach demonstrates the potential for improving respiratory disorders diagnosis and hence leading to improved patient care.

AUTOMATED HATE SPEECH DETECTION USING RECURRENT NEURAL NETWORK

The rising use of online amusement and information sharing has given critical promotion advantages to humankind. Regardless, this has in like manner prompted various troubles including the spreading likewise, sharing of scorn talk messages. Thus, to tackle this emerging issue in virtual diversion objections, continuous assessments utilized a combination of incorporated planning techniques and artificial intelligence estimations to distinguish the contempt talk messages on different datasets. Nevertheless, suppose, there is no audit to investigate the variety of component planning strategies and Artificial intelligence computations to evaluate which incorporate planning strategy and simulated intelligence estimation beat on a standard transparently open dataset. In the published paper we are going to distinguish discourse utilizing repetitive brain organization. In this system, we get the 87.38% accuracy for 300 epochs. Keywords— Convolutional Neural Network, Deep Learning, Recurrent Neural Network

Challenges as catalysts: how Waymo’s Open Dataset Challenges shape AI development

AbstractArtificial intelligence (AI) and machine learning (ML) are becoming increasingly significant areas of research for scholars in science and technology studies (STS) and media studies. In March 2020, Waymo, Google/Alphabet’s autonomous vehicle project, introduced the ‘Open Dataset Virtual Challenge’, an annual competition leveraging their Waymo Open Dataset. This freely accessible dataset comprises annotated autonomous vehicle data from their own Waymo vehicles. Yearly, Waymo has continued to host iterations of this challenge, inviting teams of computer scientists to tackle evolving machine learning and vision problems using Google's data and tools. This article analyses these challenges, situating them within the context of the ‘Grand Challenges’ of artificial intelligence (AI), which aimed to foster accountable and commercially viable advancements in the late 1980s. Through two exploratory workshops, we adopted a ‘technographic’ approach to examine the pivotal role of challenges in the development and political economy of AI. Serving as an organising principle for the AI innovation ecosystem, the challenge connects companies and external collaborators, driving advancements in specific machine vision domains. By exploring six key themes—interface methods, incrementalism, metrics, AI vernacular, applied domains, and competitive advantages—the article illustrates the role of these challenges in shaping AI research and development. By unpacking the dynamic interaction between data, computation, and labour, these challenges serve as catalysts propelling advancements towards self-driving technologies. The study reveals how challenges have historically and presently shaped the evolving landscape of self-driving and AI technologies.

A disparity‐aware Siamese network for building change detection in bi‐temporal remote sensing images

AbstractBuilding change detection has various applications, such as urban management and disaster assessment. Along with the exponential growth of remote sensing data and computing power, an increasing number of deep‐learning‐based remote sensing building change detection methods have been proposed in recent years. Objectively, the overwhelming majority of existing methods can perfectly deal with the change detection of low‐rise buildings. By contrast, high‐rise buildings often present a large disparity in multitemporal high‐resolution remote sensing images, which degrades the performance of existing methods dramatically. To alleviate this problem, we propose a disparity‐aware Siamese network for detecting building changes in bi‐temporal high‐resolution remote sensing images. The proposed network utilises a cycle‐alignment module to address the disparity problem at both the image and feature levels. A multi‐task learning framework with joint semantic segmentation and change detection loss is used to train the entire deep network, including the cycle‐alignment module in an end‐to‐end manner. Extensive experiments on three publicly open building change detection datasets demonstrate that our method achieves significant improvements on datasets with severe building disparity and state‐of‐the‐art performance on datasets with minimal building disparity simultaneously.

Meditation-type specific reduction in infra-slow activity of electroencephalogram.

Meditation is renowned for its positive effects on cognitive abilities and stress reduction. It has been reported that the amplitude of electroencephalographic (EEG) infra-slow activity (ISA, < 0.1Hz) is reduced as the stress level decreases. Consequently, we aimed to determine if EEG ISA amplitude decreases as a result of meditation practice across various traditions. To this end, we analyzed an open dataset comprising EEG data acquired during meditation sessions from experienced practitioners in the Vipassana tradition-which integrates elements of focused attention and open monitoring, akin to mindfulness meditation-and in the Himalayan Yoga and Isha Shoonya traditions, which emphasize focused attention and open monitoring, respectively. A general trend was observed where EEG ISA amplitude tended to decrease in experienced meditators from these traditions compared to novices, particularly significant in the 0.03-0.08Hz band for Vipassana meditators. Therefore, our analysis focused on this ISA frequency band. Specifically, a notable decrease in EEG ISA amplitude was observed in Vipassana meditators, predominantly in the left-frontal region. This reduction in EEG ISA amplitude was also accompanied by a decrease in phase-amplitude coupling (PAC) between the ISA phase and alpha band (8-12Hz) amplitude, which implied decreased neural excitability fluctuations. Our findings suggest that not only does EEG ISA amplitude decrease in experienced meditators from traditions that incorporate both focused attention and open monitoring, but this decrease may also signify a diminished influence of neural excitability fluctuations attributed to ISA.

DP2Net: A discontinuous physical property-constrained single-source domain generalization network for tool wear state recognition

Cross-conditions tool wear monitoring has a wide application prospect in manufacturing. However, the data distribution discrepancies caused by the inconsistency of process elements restrict the generalization of models under cross-conditions or even similar conditions. The existing methods based on diversity enhancement make it difficult to effectively establish the correlation between the source domain and target domains, which limits the improvement of model generalization. Therefore, this paper details the cause of data distribution discrepancies and proposes a discontinuous physical property-constrained single-source domain generalization network for milling tool wear monitoring. Firstly, a spatial attention mechanism is introduced to weight key signal segments adaptively. Secondly, the generation module is constrained by a standard sample and is used to generate diverse samples with physical properties. Thirdly, extensive recognition experiments on an open dataset and machining experiments with three distribution discrepancy levels were conducted to verify the effectiveness of the proposed method. Finally, feature visualizations provide consistency and interpretability.

Automatic breathing phase identification based on the second derivative of the recorded lung sounds

This paper presents a new method for automatic identification of the inspiratory and expiratory breathing phases in lung sound recordings. Adventitious lung sounds (wheezes and crackles), superimposed on the breath sounds, are generally an early indication of the disease, and their timing in the breathing cycle (early/mid/late inspiratory or expiratory) has clinical significance for monitoring or diagnosing disease. Therefore, the identification of the phases of the breathing cycle is an essential step for clinical interpretation of pulmonary auscultation. The proposed algorithm is designed to be robust in the presence of adventitious lung sounds or where the breath sounds may be noisy compared to healthy lung sounds. The algorithm uses the Savitzky & Golay (SG) filter to estimate the second derivative of the lung sound signal then calculates its normalized absolute value. A threshold value is used to clip any large amplitude peaks and, following low-pas filtering, the breathing phases are visible in the plotted signal. A rule-based approach to locating peaks and troughs is then used to identify inspirations and expirations. The performance of the method is evaluated using four different datasets: (a) a longitudinal dataset recorded from 19 subjects with a diagnosis of idiopathic pulmonary fibrosis (IPF), (b) a cross-sectional dataset recorded from 55 subjects who were referred for high-resolution computed tomography (HRCT) scan of the chest for various clinical indications, (c) a longitudinal dataset recorded from 10 healthy subjects, and (d) an open access lung sounds dataset containing recordings from 41 subjects with wheeze (9 with chronic obstructive pulmonary fibrosis and 32 with asthma). On average for inspiratory phase identification the algorithm had a sensitivity of (mean (standard deviation)) 92.84 (12.88)%, positive predictive value of 90.64(15.96)%, and F1-score of 90.67 (12.53)%. For identification of the expiratory phase, the algorithm had an average sensitivity of 92.19 (13.44)%, an average positive predictive value of 91.55 (15.23)%, and an average F1-score of 90.89 (12.45)%. The method shows good potential for automatic identification of breathing phases in recorded lung sounds.

Human Activity Recognition Based on Deep Learning and Micro-Doppler Radar Data.

Activity recognition is one of the significant technologies accompanying the development of the Internet of Things (IoT). It can help in recording daily life activities or reporting emergencies, thus improving the user's quality of life and safety, and even easing the workload of caregivers. This study proposes a human activity recognition (HAR) system based on activity data obtained via the micro-Doppler effect, combining a two-stream one-dimensional convolutional neural network (1D-CNN) with a bidirectional gated recurrent unit (BiGRU). Initially, radar sensor data are used to generate information related to time and frequency responses using short-time Fourier transform (STFT). Subsequently, the magnitudes and phase values are calculated and fed into the 1D-CNN and Bi-GRU models to extract spatial and temporal features for subsequent model training and activity recognition. Additionally, we propose a simple cross-channel operation (CCO) to facilitate the exchange of magnitude and phase features between parallel convolutional layers. An open dataset collected through radar, named Rad-HAR, is employed for model training and performance evaluation. Experimental results demonstrate that the proposed 1D-CNN+CCO-BiGRU model demonstrated superior performance, achieving an impressive accuracy rate of 98.2%. This outperformance of existing systems with the radar sensor underscores the proposed model's potential applicability in real-world scenarios, marking a significant advancement in the field of HAR within the IoT framework.

A dataset for soil organic carbon in agricultural systems for the Southeast Asia region

The determination of changes in soil organic carbon (SOC) content under different cropping systems is necessary for policy development oriented towards soil conservation, C sequestration, and future C credit markets. The aim of this study was to generate an open SOC dataset resulting from a systematic literature search related to the agricultural systems for Southeast Asia. The dataset has 209 articles and 4341 observations on soils of cropping systems in this region from articles published between 1987 and 2023. This dataset included different management practices, land uses, soil sampling depth, and length of SOC content assessment. In addition, inherent features of crop production reported in the experiments were included in the dataset. This dataset can be applied to quantify and compare the impact of different land uses or management practices on SOC content, providing foundational knowledge towards identifying sustainable practices. Lastly, it is a useful guide for future regional SOC sequestration policies and the development of C credit markets.

A hybrid knowledge graph for efficient exploration of lithostratigraphic information in open text data

Rocks formed during different geologic time record the diverse evolution of the geosphere and biosphere. In the past decades, substantial geoscience data have been made open access, providing invaluable resources for studying the stratigraphy in different regions and at different scales. However, many open datasets have information recorded in natural language with heterogeneous terminologies, short of efficient approaches to analyze them. In this research, we constructed a hybrid Stratigraphic Knowledge Graph (StraKG) to help address this challenge. StraKG has two layers, a simple schema layer and a rich instance layer. For the schemas, we used a short but functional list of classes and relationships, and then incorporated community-recognized terminologies from geological dictionaries. For the instances, we used natural language processing techniques to analyze open text data and obtained massive records, such as rocks and spatial locations. The nodes in the two layers were associated to establish a consistent structure of stratigraphic knowledge. To verify the functionality of StraKG, we applied it to the Baidu encyclopedia, the largest online Chinese encyclopedia. Three experiments were implemented on the topics of stratigraphic correlation, spatial distribution of ophiolite in China, and spatio-temporal distribution of open lithostratigraphic data. The results show that StraKG can provide strong knowledge reference for stratigraphic studies. Used together with data exploration and data mining methods, StraKG illustrates a new approach to analyze the open and big text data in geoscience.

Assessment of fitness for open repair in patients with infrarenal abdominal aortic aneurysms

ObjectiveEndovascular aortic repair (EVAR) was originally designed as a treatment modality for patients with abdominal aortic aneurysms (AAAs) deemed unfit for open repair. However, the definition of “unfit for open repair” is largely subjective and heterogenous. The purpose of this study was to compare patients deemed unfit for open repair who underwent EVAR to a matched cohort who underwent open repair for infrarenal AAAs. MethodsThe Vascular Quality Initiative of the Society for Vascular Surgery was queried for patients who underwent EVAR and open infrarenal AAA repair from 2003 to 2022. Patients that underwent EVAR were included if they were deemed unfit for open repair by the operating surgeon. EVAR patients deemed unfit because of a hostile abdomen were excluded. Patients in both the open and EVAR datasets were excluded if their repair was deemed non-elective or if they had prior aortic surgery. EVAR patients were matched to a cohort of open patients. The primary outcome for this study was 1-year mortality. Secondary outcomes included 30-day mortality, major adverse cardiac events, pulmonary complications, non-home discharge, reinterventions, and 5-year survival. ResultsA total of 5310 EVAR patients were identified who were deemed unfit for open repair. Of those, 3028 EVAR patients (57.0%) were able to be matched 1:1 to a cohort of open patients. Open patients had higher rates of major adverse cardiac events (20.2% vs 4.4%; P < .001), pulmonary complications (12.8% vs 1.6%; P < .001), non-home discharges (28.5% vs 7.9%; P < .001), and 30-day mortality (4.5% vs 1.4%; P < .001). There were no differences in early survival, but open repair had better middle and late survival compared with EVAR over the course of 5 years. A total of 74 EVAR patients (2.4%) had reinterventions during the study period. EVAR patients that required interventions had higher 1-year (40.5% vs 7.3%; P < .001) and 5-year mortality (43.2% vs 14.1%; P < .001) compared with those that did not require reinterventions. EVAR patients who had reinterventions had higher 1-year (40.5% vs 6.3%; P < .001) and 5-year (43.2% vs 20.3%; P = .006) mortality compared with their matched open cohort. ConclusionsPatients undergoing EVAR for AAAs who are deemed unfit for open repair have better perioperative morbidity and mortality compared with open repair. However, patients who had an open repair had better middle and late survival over the course of 5 years. The categorization of unfitness for open surgery may be inaccurate and re-evaluation of this terminology/concept should be undertaken.

Integrated Spatial and Temporal Features Based Network Intrusion Detection System Using SMOTE Sampling

With attackers discovering more inventive ways to take advantage of network weaknesses, the pace of attacks has drastically increased in recent years. As a result, network security has never been more important, and many network intrusion detection systems (NIDS) rely on old, out-of-date attack signatures. This necessitates the deployment of reliable and modern Network Intrusion Detection Systems that are educated on the most recent data and employ deep learning techniques to detect malicious activities. However, it has been found that the most recent datasets readily available contain a large quantity of benign data, enabling conventional deep learning systems to train on the imbalance data. A high false detection rate result from this. To overcome the aforementioned issues, we suggest a Synthetic Minority Over-Sampling Technique (SMOTE) integrated convolution neural network and bi-directional long short-term memory SCNN-BIDLSTM solution for creating intrusion detection systems. By employing the SMOTE, which integrates a convolution neural network to extract spatial features and a bi-directional long short-term memory to extract temporal information; difficulties are reduced by increasing the minority samples in our dataset. In order to train and evaluate our model, we used open benchmark datasets as CIC-IDS2017, NSL-KDD, and UNSW-NB15 and compared the results with other state of the art models.

Improving Computer Vision-Based Wildfire Smoke Detection by Combining SE-ResNet with SVM

Wildfire is one of the most critical natural disasters that poses a serious threat to human lives as well as ecosystems. One issue hindering a high accuracy of computer vision-based wildfire detection is the potential for water mists and clouds to be marked as wildfire smoke due to the similar appearance in images, leading to an unacceptable high false alarm rate in real-world wildfire early warning cases. This paper proposes a novel hybrid wildfire smoke detection approach by combining the multi-layer ResNet architecture with SVM to extract the smoke image dynamic and static characteristics, respectively. The ResNet model is improved via the SE attention mechanism and fully convolutional network as SE-ResNet. A fusion decision procedure is proposed for wildfire early warning. The proposed detection method was tested on open datasets and achieved an accuracy of 98.99%. The comparisons with AlexNet, VGG-16, GoogleNet, SE-ResNet-50 and SVM further illustrate the improvements.

Exploring contactless techniques in multimodal emotion recognition: insights into diverse applications, challenges, solutions, and prospects

In recent years, emotion recognition has received significant attention, presenting a plethora of opportunities for application in diverse fields such as human–computer interaction, psychology, and neuroscience, to name a few. Although unimodal emotion recognition methods offer certain benefits, they have limited ability to encompass the full spectrum of human emotional expression. In contrast, Multimodal Emotion Recognition (MER) delivers a more holistic and detailed insight into an individual's emotional state. However, existing multimodal data collection approaches utilizing contact-based devices hinder the effective deployment of this technology. We address this issue by examining the potential of contactless data collection techniques for MER. In our tertiary review study, we highlight the unaddressed gaps in the existing body of literature on MER. Through our rigorous analysis of MER studies, we identify the modalities, specific cues, open datasets with contactless cues, and unique modality combinations. This further leads us to the formulation of a comparative schema for mapping the MER requirements of a given scenario to a specific modality combination. Subsequently, we discuss the implementation of Contactless Multimodal Emotion Recognition (CMER) systems in diverse use cases with the help of the comparative schema which serves as an evaluation blueprint. Furthermore, this paper also explores ethical and privacy considerations concerning the employment of contactless MER and proposes the key principles for addressing ethical and privacy concerns. The paper further investigates the current challenges and future prospects in the field, offering recommendations for future research and development in CMER. Our study serves as a resource for researchers and practitioners in the field of emotion recognition, as well as those intrigued by the broader outcomes of this rapidly progressing technology.

Diver-robot communication dataset for underwater hand gesture recognition

In this paper, we present a dataset of diving gesture images used for human–robot interaction underwater. By offering this open access dataset, the paper aims at investigating the potential of using visual detection of diving gestures from an autonomous underwater vehicle (AUV) as a form of communication with a human diver. In addition to the image recording, the same dataset was recorded using a smart gesture recognition glove. The glove uses dielectric elastomer sensors and on-board processing to determine the selected gesture and transmit the command associated with the gesture to the AUV via acoustics. Although this method can be used under different visibility conditions and even without line of sight, it introduces a communication delay required for the acoustic transmission of the gesture command. To compare efficiency, the glove was equipped with visual markers proposed in a gesture-based language called CADDIAN and recorded with an underwater camera in parallel to the glove’s onboard recognition process. The dataset contains over 30,000 underwater frames of nearly 900 individual gestures annotated in corresponding snippet folders. The dataset was recorded in a balanced ratio with five different divers in sea and five different divers in pool conditions, with gestures recorded at 1, 2 and 3 metres from the camera. The glove gesture recognition statistics are reported in terms of average diver reaction time, average time taken to perform a gesture, recognition success rate, transmission times and more. The dataset presented should provide a good baseline for comparing the performance of state of the art visual diving gesture recognition techniques under different visibility conditions.