Video-SafetyBench: A Benchmark for Safety Evaluation of Video-Language Understanding 🌐 Homepage | 🤗 Dataset | 📖 arXiv | GitHub Dataset Details Our dataset statistics and some examples are listed in the following:

License The Video-SafetyBench is under the CC BY-NC-SA 4.0.Ethics Statement The dataset we created is intended solely for AI safety research and learning, with the goal of assessing the safety ability of current video LVLMs. Our data collection process does not involve user… See the full description on the dataset page: https://huggingface.co/datasets/BAAI/Video-SafetyBench.

This dataset contains all comments (comments and replies) of the YouTube vision video "Tunnels" by "The Boring Company" fetched on 2020-10-13 using YouTube API. The comments are classified manually by three persons. We performed a single-class labeling of the video comments regarding their relevance for requirement engineering (RE) (ham/spam), their polarity (positive/neutral/negative). Furthermore, we performed a multi-class labeling of the comments regarding their intention (feature request and problem report) and their topic (efficiency and safety). While a comment can only be relevant or not relevant and have only one polarity, a comment can have one or more intentions and also one or more topics.

For the replies, one person also classified them regarding their relevance for RE. However, the investigation of the replies is ongoing and future work.

Remark: For 126 comments and 26 replies, we could not determine the date and time since they were no longer accessible on YouTube at the time this data set was created. In the case of a missing date and time, we inserted "NULL" in the corresponding cell.

This data set includes the following files:

Dataset.xlsx contains the raw and labeled video comments and replies:

For each comment, the data set contains:

ID: An identification number generated by YouTube for the comment

Date: The date and time of the creation of the comment

Author: The username of the author of the comment

Likes: The number of likes of the comment

Replies: The number of replies to the comment

Comment: The written comment

Relevance: Label indicating the relevance of the comment for RE (ham = relevant, spam = irrelevant)

Polarity: Label indicating the polarity of the comment

Feature request: Label indicating that the comment request a feature

Problem report: Label indicating that the comment reports a problem

Efficiency: Label indicating that the comment deals with the topic efficiency

Safety: Label indicating that the comment deals with the topic safety

For each reply, the data set contains:

ID: The identification number of the comment to which the reply belongs

Date: The date and time of the creation of the reply

Author: The username of the author of the reply

Likes: The number of likes of the reply

Comment: The written reply

Relevance: Label indicating the relevance of the reply for RE (ham = relevant, spam = irrelevant)

Detailed analysis results.xlsx contains the detailed results of all ten times repeated 10-fold cross validation analyses for each of all considered combinations of machine learning algorithms and features

Guide Sheet - Multi-class labeling.pdf describes the coding task, defines the categories, and lists examples to reduce inconsistencies and increase the quality of manual multi-class labeling

Guide Sheet - Single-class labeling.pdf describes the coding task, defines the categories, and lists examples to reduce inconsistencies and increase the quality of manual single-class labeling

Python scripts for analysis.zip contains the scripts (as jupyter notebooks) and prepared data (as csv-files) for the analyses

Disclaimer - The calls listed here are only those where the element assigned to the call has arrived and is currently working the call. It does not include any calls for service, whether currently being worked or not, that are not releasable due to privacy laws.

Data and replication codes for the article "Fostering safe food handling among consumers: Causal evidence on game- and video-based online interventions". 1,973 participants from the UK and Norway, aged 18- 89 years, were assigned to (i) a control condition, or (ii) exposed to a brief information video, or (iii) in addition played an online game (two different conditions). In all conditions, participants answered a pre-survey and seven days later a post-survey. In the survey, next to collecting some information on sociodemographic background and certain preferences, subjects reported some recent food safety behaviors and we elicited beliefs in the efficacy of certain food safety actions, as well as beliefs in myths related to food and hygiene.

We use this data set in our publication
Koch, A. K., Mønster, D., Nafziger, J., & Veflen, N. (2022). Fostering safe food handling among consumers: Causal evidence on game-and video-based online interventions. Food Control, 108825.

Synthetic Data Video Generator Market Outlook

According to our latest research, the global synthetic data video generator market size reached USD 1.32 billion in 2024 and is anticipated to grow at a robust CAGR of 38.7% from 2025 to 2033. By the end of 2033, the market is projected to reach USD 18.59 billion, driven by rapid advancements in artificial intelligence, the growing need for high-quality training data for machine learning models, and increasing adoption across industries such as autonomous vehicles, healthcare, and surveillance. The surge in demand for data privacy, coupled with the necessity to overcome data scarcity and bias in real-world datasets, is significantly fueling the synthetic data video generator market's growth trajectory.

One of the primary growth factors for the synthetic data video generator market is the escalating demand for high-fidelity, annotated video datasets required to train and validate AI-driven systems. Traditional data collection methods are often hampered by privacy concerns, high costs, and the sheer complexity of obtaining diverse and representative video samples. Synthetic data video generators address these challenges by enabling the creation of large-scale, customizable, and bias-free datasets that closely mimic real-world scenarios. This capability is particularly vital for sectors such as autonomous vehicles and robotics, where the accuracy and safety of AI models depend heavily on the quality and variety of training data. As organizations strive to accelerate innovation and reduce the risks associated with real-world data collection, the adoption of synthetic data video generation technologies is expected to expand rapidly.

Another significant driver for the synthetic data video generator market is the increasing regulatory scrutiny surrounding data privacy and compliance. With stricter regulations such as GDPR and CCPA coming into force, organizations face mounting challenges in using real-world video data that may contain personally identifiable information. Synthetic data offers an effective solution by generating video datasets devoid of any real individuals, thereby ensuring compliance while still enabling advanced analytics and machine learning. Moreover, synthetic data video generators empower businesses to simulate rare or hazardous events that are difficult or unethical to capture in real life, further enhancing model robustness and preparedness. This advantage is particularly pronounced in healthcare, surveillance, and automotive industries, where data privacy and safety are paramount.

Technological advancements and increasing integration with cloud-based platforms are also propelling the synthetic data video generator market forward. The proliferation of cloud computing has made it easier for organizations of all sizes to access scalable synthetic data generation tools without significant upfront investments in hardware or infrastructure. Furthermore, the continuous evolution of generative adversarial networks (GANs) and other deep learning techniques has dramatically improved the realism and utility of synthetic video data. As a result, companies are now able to generate highly realistic, scenario-specific video datasets at scale, reducing both the time and cost required for AI development. This democratization of synthetic data technology is expected to unlock new opportunities across a wide array of applications, from entertainment content production to advanced surveillance systems.

From a regional perspective, North America currently dominates the synthetic data video generator market, accounting for the largest share in 2024, followed closely by Europe and Asia Pacific. The strong presence of leading AI technology providers, robust investment in research and development, and early adoption by automotive and healthcare sectors are key contributors to North America's market leadership. Europe is also witnessing significant growth, driven by stringent data privacy regulations and increased focus on AI-driven innovation. Meanwhile, Asia Pacific is emerging as a high-growth region, fueled by rapid digital transformation, expanding IT infrastructure, and increasing investments in autonomous systems and smart city projects. Latin America and Middle East & Africa, while still nascent, are expected to experience steady uptake as awareness and technological capabilities continue to grow.

Component Analysis

The synthetic data video generator market by comp

Lending library for King County employees only

40 People – Safety Dressing Collection Data. Each subject collects 24 videos, each video lasts about 30 seconds. The gender distribution includes male and female, the age distribution is young and middle-aged. Collecting scenes include 2 indoor scenes and 2 outdoor scenes. The collecting angles are looking down angle, looking up angle. The data diversity includes multiple scenes, multiple actions, multiple angles, multiple safety dressing equipment. The data can be used for tasks such as detection and recognition of safety dressing for power personnel.

Have you ever wondered if vaccines are safe? Watch this video to follow the journey that every vaccine goes on before it is used in Canada.

At Driver Technologies, we specialize in collecting high-quality, highly-anonymized driving data crowdsourced through our dash cam app. Our Motorcycle Machine Learning Video Data is built from millions of miles of driving data captured by our users and is optimized for training machine learning models, enhancing various applications in road safety, and advancing the future of mobility.

What Makes Our Data Unique? What sets our Motorcycle Machine Learning Video Data apart is its comprehensive approach to road object detection. By leveraging advanced machine learning models, we analyze the captured video to identify and classify various road objects encountered during a trip. This includes vehicles, pedestrians, traffic signs androad conditions, resulting in rich, annotated datasets that can be applied across a wide range of applications.

How Is the Data Generally Sourced? Our data is sourced directly from users who utilize our dash cam app, which harnesses the smartphone’s camera and sensors to record during a trip. This direct sourcing method ensures that our data is unbiased and represents a wide variety of conditions and environments. The data is not only authentic and reflective of current road conditions but is also abundant in volume, offering millions of miles of recorded trips that cover diverse scenarios.

Primary Use-Cases and Verticals The Motorcycle Machine Learning Video Data is tailored for various sectors, particularly those involved in motorcycle safety, transportation planning, and autonomous vehicle development. Key use cases include:

Training Machine Learning Models: Clients can utilize our annotated data to develop and refine machine learning models for applications in road safety and autonomous vehicle systems, ensuring better object detection and decision-making capabilities.

Urban Planning and Infrastructure Development: Our data helps municipalities understand road usage patterns, particularly those involving motorcycles, enabling them to make informed decisions regarding infrastructure improvements, safety measures, and traffic management.

Insurance Analytics: Insurance companies can leverage insights from our data to assess risk in various environments, aiding in the development of tailored insurance products for motorcyclists and improving claims processing.

Integration with Our Broader Data Offering The Motorcycle Machine Learning Video Data is an essential component of our broader data offerings at Driver Technologies. It complements our extensive library of driving data collected from various vehicles and road users, creating a comprehensive data ecosystem that supports multiple verticals, including insurance, automotive technology, and machine learning models.

In summary, Driver Technologies' Motorcycle Machine Learning Video Data provides a unique opportunity for data buyers to access high-quality, actionable insights that drive innovation in road safety and mobility. By integrating our Motorcycle Machine Learning Video Data with other datasets, clients can gain a holistic view of transportation dynamics, enhancing their analytical capabilities and decision-making processes.

This dataset was collected as part of the U.S. Department of Transportation (U.S. DOT) Intersection Safety Challenge (hereafter, “the Challenge”) for Stage 1B: System Assessment and Virtual Testing. Multi-sensor data were collected at a controlled test roadway intersection the Federal Highway Administration (FHWA) Turner-Fairbank Highway Research Center (TFHRC) Smart Intersection facility in McLean, VA from October 2023 through March 2024. The data include potential conflict-based and non-conflict-based experimental scenarios between vulnerable road users (e.g., pedestrians, bicyclists) and vehicles during both daytime and nighttime conditions. Note that no actual human vulnerable road users were put at risk of being involved in a collision during the data collection efforts. The provided data (hereafter, “the Challenge Dataset”) are unlabeled training data (without ground truth) that were collected to be used for intersection safety system algorithm training, refinement, tuning, and/or validation, but may have additional uses. For a summary of the Stage 1B data collection effort, please see this video: https://youtu.be/csirVHFa2Cc. The Challenge Dataset includes data at a single, signalized four-way intersection from 20 roadside sensors and traffic control devices, including eight closed-circuit television (CCTV) visual cameras, five thermal cameras, two light detection and ranging (LiDAR) sensors, and four radar sensors. Intrinsic calibration was performed for all visual and thermal cameras. Extrinsic calibration was performed for specific pairs of roadside sensors. Additionally, the traffic signal phase and timing data and vehicle and/or pedestrian calls to the traffic signal controller (if any) are also provided. The total number of unique runs in the Challenge Dataset is 1,104, bringing the total size of the dataset to approximately 1 TB. A sample of 20 unique runs from the Challenge Dataset is provided here for download, inspection, and use. If, after inspecting this sample, a potential data user would like access to download the full Challenge Dataset, a request can be made via the form here: https://its.dot.gov/data/data-request For more details about the data collection, supplemental files, organization and dictionary, and sensor calibration, see the attached “U.S. DOT ISC Stage 1B ITS DataHub Metadata_v1.0.pdf” document. For more information on the background of the Intersection Safety Challenge Stage 1B, please visit: https://www.its.dot.gov/research-areas/Intersection-Safety-Challenge/.

The Africa Data Center Physical Security Market is experiencing robust growth, projected to reach $42.69 million in 2025 and maintain a Compound Annual Growth Rate (CAGR) of 19.10% from 2025 to 2033. This expansion is driven by several factors. The increasing adoption of cloud computing and the burgeoning digital economy across Africa necessitate robust data center security infrastructure. Government initiatives promoting digital transformation and investments in critical national infrastructure are further fueling demand. The rising prevalence of cyber threats and data breaches compels organizations to prioritize physical security measures, such as video surveillance, access control systems, and robust perimeter protection. Furthermore, the growing need for reliable and secure data storage across various sectors, including IT & Telecommunications, BFSI (Banking, Financial Services, and Insurance), government, and healthcare, is a significant driver. Key players like Bosch, Axis Communications, Siemens, and Johnson Controls are actively shaping the market landscape with their advanced solutions. The market segmentation, encompassing various solution types (video surveillance, access control, etc.), service types (consulting, professional services, system integration), and end-users, highlights the diversity of opportunities within this dynamic market. Focus on specialized services like system integration and professional services is likely to increase as more complex security needs arise. The significant growth is expected across major African economies such as Nigeria, South Africa, Egypt, Kenya, and others, reflecting the continent’s overall technological progress and digitalization. The market's future trajectory suggests a continued upward trend. The ongoing investment in 5G infrastructure and the expanding reach of internet connectivity are expected to further amplify the demand for secure data centers. However, challenges such as infrastructural limitations in certain regions, varying levels of cybersecurity awareness, and the initial high capital expenditure associated with advanced security systems could potentially moderate growth. Nonetheless, the overall market outlook remains highly positive, driven by the imperative need for secure data storage and the continued growth of the African digital economy. This translates into significant opportunities for both established players and new entrants to participate in this expanding market. Recent developments include: October 2023: Zwipe partnered with Schneider Electric’s Security Solutions Group. Schneider Electric will introduce its clientele to the Zwipe Access fingerprint-scanning smart card. This card will be integrated with Schneider Electric’s Continuum and Security Expert platforms, serving a client base from sectors, including airports, transportation, healthcare, and data centers., April 2023: Schneider Electric launched a new service offer, EcoCare for Modular Data Centers services membership. Members of this innovative service plan benefit from specialized expertise to maximize modular data centers' uptime with 24/7 proactive remote monitoring and condition-based maintenance.. Key drivers for this market are: Growing Adoption of Access Control Systems Owing to Rising Crime Rates and Threats, Advancements in Video Surveillance Systems Connected to Cloud Systems. Potential restraints include: Growing Adoption of Access Control Systems Owing to Rising Crime Rates and Threats, Advancements in Video Surveillance Systems Connected to Cloud Systems. Notable trends are: The IT and Telecom Segment to Hold Significant Share.

Seattle Police Department In-Car Video Dropped Frame Report

The ASAPS Dataset includes eight continuous hours of data from across a fictitious ASAPS City representative of a normal day in a small city. The dataset includes multiple data types - video, audio, text, sensor, social media. The video data was created from a series of staged events, and then synchronized and augmented with recorded audio, simulated data for sensor, text, and social media related to the variety of emergency events across the city. The dataset represents a geographic area of approximately 68 square blocks. All of the GPS coordinates represent a fictitious location which includes 390 buildings and parks, each with an identity/name, street address, and latitude/longitude location. The first-of-its-kind dataset combines video recordings of staged emergencies with scripted audio and textual public safety communications and simulated Computer Aided Dispatch and gunshot sensor data. It was constructed to represent a continuous eight-hour snapshot of the emergencies happening during the day in a small city and includes 45 time-synchronized data streams. No actual emergencies occur in the data and it includes only simulated data and audio and video recordings of consenting subjects who participated as actors in the staging of the data. The subjects agreed to their video and audio being recorded and distributed to support open research in automated emergency analysis in a formal data collection consent process. The ASAPS Dataset was created by the Lafayette Group INC under contract award #GS-23F-0134N on behalf of NIST PSCR. The data collection protocol was reviewed and approved in April 2020 by the New England Institutional Review Board (NEIRB) in accordance with 45 CFR 46, the Protection of Human Subjects and by the NIST Research Protections Office in accordance with 15 CFR 27.112, Review by Institution. The data is organized and annotated to be used for R&D in automated emergency event analysis and its use is restricted to the terms and conditions of the data use agreement. The ASAPS Dataset is available to Data Recipients by registering for an account via the ASAPS Dataset website (https://asapsdata.nist.gov).

The chart shows the number of videos removed by Google under their Child Safety policy, starting from September 2018. The latest available data shows that overall the number of videos removed under the Child Safety Policy declined in in the first quarter of 2022 by 81% compared to the same period of the previous year. Compared to the previous quarter, the change is considerably lower, declining only by 18% in the first quarter of 2022 compared to the previous one.

The global Video Event Data Recorder (VEDR) market is poised for significant growth, projected to reach $[Estimated 2025 Market Size in Millions] million by 2025 and $[Estimated 2033 Market Size in Millions] million by 2033, exhibiting a Compound Annual Growth Rate (CAGR) of 5.2%. This expansion is driven by several key factors. The increasing adoption of advanced driver-assistance systems (ADAS) and the growing demand for enhanced fleet safety and management solutions are primary contributors. Furthermore, stringent government regulations regarding commercial vehicle safety and the rising need for evidence-based accident reconstruction are fueling market growth. Technological advancements, such as the integration of AI and machine learning for improved video analytics and data processing, are also playing a crucial role. Leading players like Digital Ally, Octo Telematics, and WatchGuard Video are constantly innovating, introducing sophisticated VEDR solutions with enhanced features and functionalities, further driving market penetration. The market is segmented by vehicle type (commercial, passenger), recording technology (HD, 4K), and application (fleet management, law enforcement). While initial investment costs can be a restraint, the long-term benefits of improved safety, reduced insurance premiums, and enhanced operational efficiency are compelling businesses to adopt VEDR technology. The competitive landscape is characterized by both established players and emerging technology providers. Established players benefit from extensive distribution networks and strong brand recognition, while newer companies are focusing on providing innovative solutions at competitive price points. Regional variations in market growth are likely, with North America and Europe expected to be leading markets, driven by high vehicle density, stringent regulations, and high adoption rates in fleet management. However, growth opportunities also exist in developing economies in Asia-Pacific and Latin America, as transportation infrastructure and safety standards improve. The future of the VEDR market hinges on the continued development of AI-powered analytics, integration with connected vehicle technologies, and the increasing adoption of cloud-based data storage and management solutions.

Creating a strong password.

Europe Video Surveillance Market size was estimated at USD 8.99 Billion in 2024 and is projected to reach USD 19.93 Billion by 2031, growing at a CAGR of 10.46% from 2024 to 2031.Europe Video Surveillance Market DriversTechnological Progress1. Integration of AI and Machine Learning:The market is being revolutionized by the integration of machine learning (ML) and artificial intelligence (AI) into video surveillance systems. These technologies allow for real-time analytics, object detection, facial recognition, and behavior analysis, which improves the capabilities of surveillance systems. AI-driven solutions increase the efficacy and efficiency of security measures by anticipating and thwarting possible threats.2. High-Resolution Imaging: There is an increasing need for cameras with high resolution, such as 4K and even 8K. For precise identification and evidence gathering, these cameras produce sharper, more detailed images. Improved imaging technologies guarantee improved performance under different lighting scenarios, which propels market expansion even further.3. Cloud-Based Solutions: The adoption of cloud-based video surveillance systems provides better data management, scalability, and remote access. Large volumes of video data may be readily stored and retrieved with cloud storage, enabling real-time monitoring and speedy footage retrieval. The increasing demand for streamlined and consolidated surveillance systems is supported by this shift.Normative Structures1. GDPR Compliance: Stricter privacy and data protection requirements have been imposed throughout Europe by the General Data Protection Regulation (GDPR). These rules must be followed by video surveillance systems in order to guarantee responsible and safe data handling. Investments in compliant surveillance procedures and technologies have surged as a result.2. Public Safety Initiatives: Numerous public safety initiatives, such as city-wide surveillance programs, are being carried out by governments and municipal authorities. These programs seek to lower crime rates, improve urban safety, and improve public space management. The market is significantly driven by government financing and regulations that assist the development of smart cities.

• The Ministry of the Interior and Safety publishes the 'Administrative Safety Statistical Yearbook' every year by compiling statistical data from the headquarters of the Ministry of the Interior and Safety and its affiliated organizations in accordance with the 'Ministry of the Interior and Safety Statistics Management Regulations'. - The statistical information by field included in the 'Administrative Safety Statistical Yearbook' is provided as an open API so that it can be used in various fields in both the public and private sectors. - The open API in question is statistics on 'PC video conferencing utilization' among the 'digital government' included in the 'Administrative Safety Statistical Yearbook'. It provides statistical information on the number of PC video conferencing participants by organization (central government, local government) and year. - In addition, the 'Administrative Safety Statistical Yearbook' can be downloaded in PDF format from the Ministry of the Interior and Safety website at Policy Data > Statistics > Statistical Yearbook/Statistics by Subject.

Tricks to not get sick: Food safety for kids (described video)

This dataset includes Police In-Car video activities. Data shows activity of In-Car recorded video including, Officer Serial Number, Start (Date & Time) of video, post recording activity code and description. Data is refreshed on a weekly basis. Data is records of video and does not imply that there is actual playable video associated.