Number and percentage of homicide victims, by type of firearm used to commit the homicide (total firearms; handgun; rifle or shotgun; fully automatic firearm; sawed-off rifle or shotgun; firearm-like weapons; other firearms, type unknown), Canada, 1974 to 2018.

THIS DATASET WAS LAST UPDATED AT 2:11 PM EASTERN ON JUNE 25

OVERVIEW

2019 had the most mass killings since at least the 1970s, according to the Associated Press/USA TODAY/Northeastern University Mass Killings Database.

In all, there were 45 mass killings, defined as when four or more people are killed excluding the perpetrator. Of those, 33 were mass shootings . This summer was especially violent, with three high-profile public mass shootings occurring in the span of just four weeks, leaving 38 killed and 66 injured.

A total of 229 people died in mass killings in 2019.

The AP's analysis found that more than 50% of the incidents were family annihilations, which is similar to prior years. Although they are far less common, the 9 public mass shootings during the year were the most deadly type of mass murder, resulting in 73 people's deaths, not including the assailants.

One-third of the offenders died at the scene of the killing or soon after, half from suicides.

About this Dataset

The Associated Press/USA TODAY/Northeastern University Mass Killings database tracks all U.S. homicides since 2006 involving four or more people killed (not including the offender) over a short period of time (24 hours) regardless of weapon, location, victim-offender relationship or motive. The database includes information on these and other characteristics concerning the incidents, offenders, and victims.

The AP/USA TODAY/Northeastern database represents the most complete tracking of mass murders by the above definition currently available. Other efforts, such as the Gun Violence Archive or Everytown for Gun Safety may include events that do not meet our criteria, but a review of these sites and others indicates that this database contains every event that matches the definition, including some not tracked by other organizations.

This data will be updated periodically and can be used as an ongoing resource to help cover these events.

Using this Dataset

To get basic counts of incidents of mass killings and mass shootings by year nationwide, use these queries:

Mass killings by year

Mass shootings by year

To get these counts just for your state:

Filter killings by state

Definition of "mass murder"

Mass murder is defined as the intentional killing of four or more victims by any means within a 24-hour period, excluding the deaths of unborn children and the offender(s). The standard of four or more dead was initially set by the FBI.

This definition does not exclude cases based on method (e.g., shootings only), type or motivation (e.g., public only), victim-offender relationship (e.g., strangers only), or number of locations (e.g., one). The time frame of 24 hours was chosen to eliminate conflation with spree killers, who kill multiple victims in quick succession in different locations or incidents, and to satisfy the traditional requirement of occurring in a “single incident.”

Offenders who commit mass murder during a spree (before or after committing additional homicides) are included in the database, and all victims within seven days of the mass murder are included in the victim count. Negligent homicides related to driving under the influence or accidental fires are excluded due to the lack of offender intent. Only incidents occurring within the 50 states and Washington D.C. are considered.

Methodology

Project researchers first identified potential incidents using the Federal Bureau of Investigation’s Supplementary Homicide Reports (SHR). Homicide incidents in the SHR were flagged as potential mass murder cases if four or more victims were reported on the same record, and the type of death was murder or non-negligent manslaughter.

Cases were subsequently verified utilizing media accounts, court documents, academic journal articles, books, and local law enforcement records obtained through Freedom of Information Act (FOIA) requests. Each data point was corroborated by multiple sources, which were compiled into a single document to assess the quality of information.

In case(s) of contradiction among sources, official law enforcement or court records were used, when available, followed by the most recent media or academic source.

Case information was subsequently compared with every other known mass murder database to ensure reliability and validity. Incidents listed in the SHR that could not be independently verified were excluded from the database.

Project researchers also conducted extensive searches for incidents not reported in the SHR during the time period, utilizing internet search engines, Lexis-Nexis, and Newspapers.com. Search terms include: [number] dead, [number] killed, [number] slain, [number] murdered, [number] homicide, mass murder, mass shooting, massacre, rampage, family killing, familicide, and arson murder. Offender, victim, and location names were also directly searched when available.

This project started at USA TODAY in 2012.

Contacts

Contact AP Data Editor Justin Myers with questions, suggestions or comments about this dataset at jmyers@ap.org. The Northeastern University researcher working with AP and USA TODAY is Professor James Alan Fox, who can be reached at j.fox@northeastern.edu or 617-416-4400.

In recent years, gun violence in the United States has become an alarmingly common occurrence. From 2016, there has been over ****** homicides by firearm in the U.S. each year and firearms have been found to make up the majority of murder weapons in the country by far, demonstrating increasing rates of gun violence occurring throughout the nation. As of 2025, Mississippi was the state with the highest gun violence rate per 100,000 residents in the United States, at **** percent, followed by Louisiana, at **** percent. In comparison, Massachusetts had a gun violence rate of *** percent, the lowest out of all the states. The importance of gun laws Gun laws in the United States vary from state to state, which has been found to affect the differing rates of gun violence throughout the country. Fewer people die by gun violence in states where gun safety laws have been passed, while gun violence rates remain high in states where gun usage is easily permitted and even encouraged. In addition, some states suffer from high rates of gun violence despite having strong gun safety laws due to gun trafficking, as traffickers can distribute firearms illegally past state lines. The right to bear arms Despite evidence from other countries demonstrating that strict gun control measures reduce rates of gun violence, the United States has remained reluctant to enact gun control laws. This can largely be attributed to the Second Amendment of the Constitution, which states that citizens have the right to bear arms. Consequently, gun control has become a highly partisan issue in the U.S., with ** percent of Democrats believing that it was more important to limit gun ownership while ** percent of Republicans felt that it was more important to protect the right of Americans to own guns.

Mortality rate from firearms includes homicides, suicides, accidental deaths, deaths by law enforcement, and deaths for which intent was undetermined. Mortality rate is based on the location of residence and has been age-adjusted to the 2000 U.S. standard population. ICD 10 codes used to identify firearm deaths are W32-W34, X72-X74, X93-X95, Y22-Y24, Y35.0, and U01.4. Single-year data are only available for Los Angeles County overall, Service Planning Areas, Supervisorial Districts, City of Los Angeles overall, and City of Los Angeles Council Districts.Violence is a public health crisis in the US, with gun violence being a major driver. In the US, the age-adjusted homicide rate from firearms is more than 20 times higher than in the European Union or in Australia. Significant disparities by age, sex, and race and ethnicity exist, with young adults (ages 15-34 years), males, and Black individuals most disproportionately impacted. Firearm-related suicides disproportionately impact older, White men. Comprehensive prevention strategies should work to address underlying physical, social, economic, and structural conditions known to increase risk.For more information about the Community Health Profiles Data Initiative, please see the initiative homepage.

The Division of Criminal Justice Services (DCJS) collects crime reports from more than 500 New York State police and sheriffs’ departments. DCJS compiles these reports as New York’s official crime statistics and submits them to the FBI under the National Uniform Crime Reporting (UCR) Program. UCR uses standard offense definitions to count crime in localities across America regardless of variations in crime laws from state to state. In New York State, law enforcement agencies use the UCR system to report their monthly crime totals to DCJS. The UCR reporting system collects information on seven crimes classified as Index offenses which are most commonly used to gauge overall crime volume. These include the violent crimes of murder/non-negligent manslaughter, forcible rape, robbery, and aggravated assault; and the property crimes of burglary, larceny, and motor vehicle theft. Firearm counts are derived from taking the number of violent crimes which involve a firearm. Population data are provided every year by the FBI, based on US Census information. Police agencies may experience reporting problems that preclude accurate or complete reporting. The counts represent only crimes reported to the police but not total crimes that occurred. DCJS posts preliminary data in the spring and final data in the fall.

Sadly, the trend of fatal police shootings in the United States seems to only be increasing, with a total 1,173 civilians having been shot, 248 of whom were Black, as of December 2024. In 2023, there were 1,164 fatal police shootings. Additionally, the rate of fatal police shootings among Black Americans was much higher than that for any other ethnicity, standing at 6.1 fatal shootings per million of the population per year between 2015 and 2024. Police brutality in the U.S. In recent years, particularly since the fatal shooting of Michael Brown in Ferguson, Missouri in 2014, police brutality has become a hot button issue in the United States. The number of homicides committed by police in the United States is often compared to those in countries such as England, where the number is significantly lower. Black Lives Matter The Black Lives Matter Movement, formed in 2013, has been a vocal part of the movement against police brutality in the U.S. by organizing “die-ins”, marches, and demonstrations in response to the killings of black men and women by police. While Black Lives Matter has become a controversial movement within the U.S., it has brought more attention to the number and frequency of police shootings of civilians.

Analysis of ‘Police Killings US’ provided by Analyst-2 (analyst-2.ai), based on source dataset retrieved from https://www.kaggle.com/azizozmen/police-killings-us on 13 February 2022.

--- Dataset description provided by original source is as follows ---

"In 2015, The Washington Post began to log every fatal shooting by an on-duty police officer in the United States. In that time there have been more than 5,000 such shootings recorded by The Post. After Michael Brown, an unarmed Black man, was killed in 2014 by police in Ferguson, Mo., a Post investigation found that the FBI undercounted fatal police shootings by more than half. This is because reporting by police departments is voluntary and many departments fail to do so. The Washington Post’s data relies primarily on news accounts, social media postings, and police reports. Analysis of more than five years of data reveals that the number and circumstances of fatal shootings and the overall demographics of the victims have remained relatively constant..." SOURCE ==> Washington Post Article

For more information about this story

This dataset has been prepared by The Washington Post (they keep updating it on runtime) with every fatal shooting in the United States by a police officer in the line of duty since Jan. 1, 2015.

2016 PoliceKillingUS DATASET
2017 PoliceKillingUS DATASET
2018 PoliceKillingUS DATASET
2019 PoliceKillingUS DATASET
2020 PoliceKillingUS DATASET

Features at the Dataset:

The file fatal-police-shootings-data.csv contains data about each fatal shooting in CSV format. The file can be downloaded at this URL. Each row has the following variables:

• id: a unique identifier for each victim
• name: the name of the victim
• date: the date of the fatal shooting in YYYY-MM-DD format
• manner_of_death: shot, shot and Tasered
• armed: indicates that the victim was armed with some sort of implement that a police officer believed could inflict harm
  • undetermined: it is not known whether or not the victim had a weapon
  • unknown: the victim was armed, but it is not known what the object was
  • unarmed: the victim was not armed
• age: the age of the victim
• gender: the gender of the victim. The Post identifies victims by the gender they identify with if reports indicate that it differs from their biological sex.
  • M: Male
  • F: Female
  • None: unknown
• race:
  • W: White, non-Hispanic
  • B: Black, non-Hispanic
  • A: Asian
  • N: Native American
  • H: Hispanic
  • O: Other
  • None: unknown
• city: the municipality where the fatal shooting took place. Note that in some cases this field may contain a county name if a more specific municipality is unavailable or unknown.
• state: two-letter postal code abbreviation
• signs of mental illness: News reports have indicated the victim had a history of mental health issues, expressed suicidal intentions or was experiencing mental distress at the time of the shooting.
• threat_level: The threat_level column was used to flag incidents for the story by Amy Brittain in October 2015. http://www.washingtonpost.com/sf/investigative/2015/10/24/on-duty-under-fire/ As described in the story, the general criteria for the attack label was that there was the most direct and immediate threat to life. That would include incidents where officers or others were shot at, threatened with a gun, attacked with other weapons or physical force, etc. The attack category is meant to flag the highest level of threat. The other and undetermined categories represent all remaining cases. Other includes many incidents where officers or others faced significant threats.
• flee: News reports have indicated the victim was moving away from officers
  • Foot
  • Car
  • Not fleeing

The threat column and the fleeing column are not necessarily related. For example, there is an incident in which the suspect is fleeing and at the same time turns to fire at gun at the officer. Also, attacks represent a status immediately before fatal shots by police while fleeing could begin slightly earlier and involve a chase. - body_camera: News reports have indicated an officer was wearing a body camera and it may have recorded some portion of the incident.

SOURCE

--- Original source retains full ownership of the source dataset ---

Number and rate (per 100,000 population) of homicide victims, Canada and Census Metropolitan Areas, 1981 to 2023.

This dataset includes the number and rate of firearm-related deaths among Virginia residents by patient health district over 5 combined years. Virginia Department of Health (VDH) health districts are assigned based on the residence of the patient at the time of death, not where the death occurred. Data include Virginia residents only, whether or not they died in Virginia. Data set includes deaths from 2018 through 2022.

The VDH Office of Vital Records tracks causes of death among Virginia residents using death certificates. Codes on the death certificate indicate underlying and contributing causes of death. Deaths are classified as firearm-related using the definition from the Centers for Disease Control and Prevention State Injury Indicators Report

The study was a comprehensive analysis of felonious killings of officers. The purposes of the study were (1) to analyze the nature and circumstances of incidents of felonious police killings and (2) to analyze trends in the numbers and rates of killings across different types of agencies and to explain these differences. For Part 1, Incident-Level Data, an incident-level database was created to capture all incidents involving the death of a police officer from 1983 through 1992. Data on officers and incidents were collected from the Law Enforcement Officers Killed and Assaulted (LEOKA) data collection as coded by the Uniform Crime Reporting (UCR) program. In addition to the UCR data, the Police Foundation also coded information from the LEOKA narratives that are not part of the computerized LEOKA database from the FBI. For Part 2, Agency-Level Data, the researchers created an agency-level database to research systematic differences among rates at which law enforcement officers had been feloniously killed from 1977 through 1992. The investigators focused on the 56 largest law enforcement agencies because of the availability of data for explanatory variables. Variables in Part 1 include year of killing, involvement of other officers, if the officer was killed with his/her own weapon, circumstances of the killing, location of fatal wounds, distance between officer and offender, if the victim was wearing body armor, if different officers were killed in the same incident, if the officer was in uniform, actions of the killer and of the officer at entry and final stage, if the killer was visible at first, if the officer thought the killer was a felon suspect, if the officer was shot at entry, and circumstances at anticipation, entry, and final stages. Demographic variables for Part 1 include victim's sex, age, race, type of assignment, rank, years of experience, agency, population group, and if the officer was working a security job. Part 2 contains variables describing the general municipal environment, such as whether the agency is located in the South, level of poverty according to a poverty index, population density, percent of population that was Hispanic or Black, and population aged 15-34 years old. Variables capturing the crime environment include the violent crime rate, property crime rate, and a gun-related crime index. Lastly, variables on the environment of the police agencies include violent and property crime arrests per 1,000 sworn officers, percentage of officers injured in assaults, and number of sworn officers.

This dataset includes the number and rate of firearm-related deaths among Virginia residents by patient health district and Sex over 5 combined years. Virginia Department of Health (VDH) health districts are assigned based on the residence of the patient at the time of death, not where the death occurred. Data include Virginia residents only, whether or not they died in Virginia. Data set includes deaths from 2018 through 2022.

The VDH Office of Vital Records tracks causes of death among Virginia residents using death certificates. Codes on the death certificate indicate underlying and contributing causes of death. Deaths are classified as firearm-related using the definition from the Centers for Disease Control and Prevention State Injury Indicators Report

Age-adjusted rate of deaths from firearms by sex, race/ethnicity, age; trends if available. Source: Santa Clara County Public Health Department, VRBIS, 2007-2016. Data as of 05/26/2017; U.S. Census Bureau; 2010 Census, Tables PCT12, PCT12H, PCT12I, PCT12J, PCT12K, PCT12L, PCT12M; generated by Baath M.; using American FactFinder; Accessed June 20, 2017. METADATA:Notes (String): Lists table title, notes and sourcesYear (String): Year of data; presented as single year or pooled years (2012-2016)Category (String): Lists the category representing the data: Santa Clara County is for total population, sex: Male and Female, race/ethnicity: African American, Asian/Pacific Islander, Latino and White (non-Hispanic White only); age categories as follows: <18, 18 to 44, 45 to 64, 65+; United States and Healthy People 2020 targetRate per 100,000 people (Numeric): Rate of deaths from firearms. Rates for age groups are reported as age-specific rates per 100,000 people. All other rates are age-adjusted rates per 100,000 people.

This dataset contains the crude number, crude rate and age-adjusted rate of firearm homicides for each U.S. county. Numbers for many counties are suppressed due to low numbers of homicides and others are flagged as unstable due to large margins of error. The CDC updates numbers annually; however, there is generally an 18 month delay in making the numbers for a given year publicly available. The homicide counts originate from death certificates from local coroner and medical examiner offices. More information is available at https://www.cdc.gov/injury/wisqars/index.html.

The Washington Post spent a year determining how many children have been affected by school shootings, beyond just those killed or injured. To do that, reporters attempted to identify every act of gunfire at a primary or secondary school during school hours since the Columbine High massacre on April 20, 1999. Using Nexis, news articles, open-source databases, law enforcement reports, information from school websites, and calls to schools and police departments, The Post reviewed more than 1,000 alleged incidents, but counted only those that happened on campuses immediately before, during or just after classes. Shootings at after-hours events, accidental discharges that caused no injuries to anyone other than the person handling the gun, and suicides that occurred privately or posed no threat to other children were excluded. Gunfire at colleges and universities, which affects young adults rather than kids, also was not counted. After finding more than 200 incidents of gun violence that met The Post’s criteria, reporters organized them in a database for analysis. Because the federal government does not track school shootings, it’s possible that the database does not contain every incident that would qualify. To calculate how many children were exposed to gunfire in each school shooting, The Post relied on enrollment figures and demographic information from the U.S. Education Department, including the Common Core of Data and the Private School Universe Survey. The analysis used attendance figures from the year of the shooting for the vast majority of the schools. Credits: Research and Reporting: John Woodrow Cox, Steven Rich and Allyson Chiu Production and Presentation: John Muyskens and Monica Ulmanu Per the terms of the Creative Commons license, CISER notes that: 1. the license for this dataset is attached as the files license.htm and license.pdf. A brief version of the Creative Commons license is also included but users should familiarize themselves with the full license before using. 2. the licensed material is located at https://github.com/washingtonpost/data-school-shootings 3. Several of the files have been modified from the format presented at the above url including creating pdf versions of the documentation files and adding SAS, Stata, and SPSS versions through the use of StatTransfer 13. 4. These adapted versions of the original files are also released through the same Creative Commons license as the original with the same license elements.

Number, percentage and rate (per 100,000 population) of homicide victims, by racialized identity group (total, by racialized identity group; racialized identity group; South Asian; Chinese; Black; Filipino; Arab; Latin American; Southeast Asian; West Asian; Korean; Japanese; other racialized identity group; multiple racialized identity; racialized identity, but racialized identity group is unknown; rest of the population; unknown racialized identity group), gender (all genders; male; female; gender unknown) and region (Canada; Atlantic region; Quebec; Ontario; Prairies region; British Columbia; territories), 2019 to 2023.

Analysis of ‘Index, Violent, Property, and Firearm Rates By County: Beginning 1990’ provided by Analyst-2 (analyst-2.ai), based on source dataset retrieved from https://catalog.data.gov/dataset/9c9619a6-a7cc-48f7-aefb-cb5c77327b9c on 27 January 2022.

--- Dataset description provided by original source is as follows ---

The Division of Criminal Justice Services (DCJS) collects crime reports from more than 500 New York State police and sheriffs’ departments. DCJS compiles these reports as New York’s official crime statistics and submits them to the FBI under the National Uniform Crime Reporting (UCR) Program. UCR uses standard offense definitions to count crime in localities across America regardless of variations in crime laws from state to state. In New York State, law enforcement agencies use the UCR system to report their monthly crime totals to DCJS. The UCR reporting system collects information on seven crimes classified as Index offenses which are most commonly used to gauge overall crime volume. These include the violent crimes of murder/non-negligent manslaughter, forcible rape, robbery, and aggravated assault; and the property crimes of burglary, larceny, and motor vehicle theft. Firearm counts are derived from taking the number of violent crimes which involve a firearm. Population data are provided every year by the FBI, based on US Census information. Police agencies may experience reporting problems that preclude accurate or complete reporting. The counts represent only crimes reported to the police but not total crimes that occurred. DCJS posts preliminary data in the spring and final data in the fall.

--- Original source retains full ownership of the source dataset ---

This is a new dataset of news headlines and their frames related to the issue of gun violence in the United States. This Gun Violence Frame Corpus (GVFC) was curated and annotated by journalism and communication experts. The articles in this dataset are drawn from a sample of news articles from a list of 30 top U.S. news websites defined in terms of traffic to the websites; and collected from four time periods over the course of 2018 in order to capture a diversity of articles.

We include in this dataset, headlines of news articles and their annotations, the accompanying images and text- and image-derived features. We also include the codebook protocol, which includes all of the variables for annotations and their definitions that are applied by the annotators.

Social determinants and the total number of firearm-related homicide incidents at the CT level in the contiguous US, 2015.

Real-world Anomaly Detection in Surveillance Videos

Surveillance videos are able to capture a variety of realistic anomalies. In this paper, we propose to learn anomalies by exploiting both normal and anomalous videos. To avoid annotating the anomalous segments or clips in training videos, which is very time consuming, we propose to learn anomaly through the deep multiple instance ranking framework by leveraging weakly labeled training videos, i.e. the training labels (anomalous or normal) are at video-level instead of clip-level. In our approach, we consider normal and anomalous videos as bags and video segments as instances in multiple instance learning (MIL), and automatically learn a deep anomaly ranking model that predicts high anomaly scores for anomalous video segments. Furthermore, we introduce sparsity and temporal smoothness constraints in the ranking loss function to better localize anomaly during training. We also introduce a new large-scale first of its kind dataset of 128 hours of videos. It consists of 1900 long and untrimmed real-world surveillance videos, with 13 realistic anomalies such as fighting, road accident, burglary, robbery, etc. as well as normal activities. This dataset can be used for two tasks. First, general anomaly detection considering all anomalies in one group and all normal activities in another group. Second, for recognizing each of 13 anomalous activities. Our experimental results show that our MIL method for anomaly detection achieves significant improvement on anomaly detection performance as compared to the state-of-the-art approaches. We provide the results of several recent deep learning baselines on anomalous activity recognition. The low recognition performance of these baselines reveals that our dataset is very challenging and opens more opportunities for future work.

Problem & Motivation

One critical task in video surveillance is detecting anomalous events such as traffic accidents, crimes or illegal activities. Generally, anomalous events rarely occur as compared to normal activities. Therefore, to alleviate the waste of labor and time, developing intelligent computer vision algorithms for automatic video anomaly detection is a pressing need. The goal of a practical anomaly detection system is to timely signal an activity that deviates normal patterns and identify the time window of the occurring anomaly. Therefore, anomaly detection can be considered as coarse level video understanding, which filters out anomalies from normal patterns. Once an anomaly is detected, it can further be categorized into one of the specific activities using classification techniques. In this work, we propose an anomaly detection algorithm using weakly labeled training videos. That is we only know the video-level labels, i.e. a video is normal or contains anomaly somewhere, but we do not know where. This is intriguing because we can easily annotate a large number of videos by only assigning video-level labels. To formulate a weakly-supervised learning approach, we resort to multiple instance learning. Specifically, we propose to learn anomaly through a deep MIL framework by treating normal and anomalous surveillance videos as bags and short segments/clips of each video as instances in a bag. Based on training videos, we automatically learn an anomaly ranking model that predicts high anomaly scores for anomalous segments in a video. During testing, a longuntrimmed video is divided into segments and fed into our deep network which assigns anomaly score for each video segment such that an anomaly can be detected.

Method

Our proposed approach (summarized in Figure 1) begins with dividing surveillance videos into a fixed number of segments during training. These segments make instances in a bag. Using both positive (anomalous) and negative (normal) bags, we train the anomaly detection model using the proposed deep MIL ranking loss. https://www.crcv.ucf.edu/projects/real-world/method.png

UCF-Crime Dataset

We construct a new large-scale dataset, called UCF-Crime, to evaluate our method. It consists of long untrimmed surveillance videos which cover 13 realworld anomalies, including Abuse, Arrest, Arson, Assault, Road Accident, Burglary, Explosion, Fighting, Robbery, Shooting, Stealing, Shoplifting, and Vandalism. These anomalies are selected because they have a significant impact on public safety. We compare our dataset with previous anomaly detection datasets in Table 1. For more details about the UCF-Crime dataset, please refer to our paper. A short description of each anomalous event is given below. Abuse: This event contains videos which show bad, cruel or violent behavior against children, old people, animals, and women. Burglary: This event contains videos that show people (thieves) entering into a building or house with the intention to commit theft. It does not include use of force against people. Robbery: This event contains videos showing thieves taking money unlawfully by force or threat of force. These videos do not include shootings. Stealing: This event contains videos showing people taking property or money without permission. They do not include shoplifting. Shooting: This event contains videos showing act of shooting someone with a gun. Shoplifting: This event contains videos showing people stealing goods from a shop while posing as a shopper. Assault: This event contains videos showing a sudden or violent physical attack on someone. Note that in these videos the person who is assaulted does not fight back. Fighting: This event contains videos displaying two are more people attacking one another. Arson: This event contains videos showing people deliberately setting fire to property. Explosion: This event contains videos showing destructive event of something blowing apart. This event does not include videos where a person intentionally sets a fire or sets off an explosion. Arrest: This event contains videos showing police arresting individuals. Road Accident: This event contains videos showing traffic accidents involving vehicles, pedestrians or cyclists. Vandalism: This event contains videos showing action involving deliberate destruction of or damage to public or private property. The term includes property damage, such as graffiti and defacement directed towards any property without permission of the owner. Normal Event: This event contains videos where no crime occurred. These videos include both indoor (such as a shopping mall) and outdoor scenes as well as day and night-time scenes. https://www.crcv.ucf.edu/projects/real-world/dataset_table.png https://www.crcv.ucf.edu/projects/real-world/method.png