Datasets / Violence Detection Dataset

Violence Detection Dataset v1 A fight is not a hug.Teach your model the difference.

Synthetic fight and violence detection dataset, with a commercial licence and no real people in the data.

Free sample and datasheet. Licence warranty: 30 days after delivery.

Annotated examples from the Violence Detection Dataset
12,540sequences
2.84 Mlabelled frames
26.3 hof events
2 to 6people per event
What it is

What is the Violence Detection Dataset?

The Xentim Violence Detection Dataset is a synthetic dataset for training skeleton-based detectors of fights in CCTV footage. It holds 12,540 sequences and 2.84 million labelled frames, with 2 to 6 people per event, multi-person keypoints, roles, actions, phases and contact events, plus hard negatives that make up about 45% of the set.

Fewer unwanted alarms.

Hugs, handshakes, high fives, dancing and crowds sit next to real violence in the same set.

Who did what to whom.

Roles, actions, phases and contact events are labelled per person, not only per clip.

A licence you can ship on.

Standard and Enterprise cover commercial use. RWF-2000, a widely used research set, needs approval from its authors first.

How it compares with RWF-2000

Xentim Violence DetectionRWF-2000
Commercial useYes, Standard and EnterpriseNot without approval from SMIIP Lab
SourceSynthetic, no real peopleReal surveillance clips
LabelsKeypoints, roles, actions, phasesFight or non-fight per clip

Sources: RWF-2000 paper. Checked 30 October 2026. Check each licence yourself before use.

What you get

Simulations1,045 events, 12 virtual cameras each
Annotationsmulti-person 2D and 3D keypoints, roles, actions, phases, contact events, camera parameters
Classes10 fight classes, 10 hard negative classes
Splitstrain, validation and test
Not includedRGB images or video, this is skeleton data

Classes

  • punch_head
  • punch_body
  • kick_body
  • kick_leg
  • shove
  • grab_strike
  • grapple
  • tackle
  • takedown
  • group_fight
  • hug
  • handshake
  • high_five
  • back_pat
  • dance_pair
  • play_push
  • heated_gesturing
  • walk_past
  • run_together
  • crowd_normal
Licences

Choose your licence.

Same dataset, different rights. Licences are perpetual.

Research

3,000 EUR net

Full dataset, same as Standard, for non-commercial research and teaching at one academic or public research institution. Results may be published. No updates or support. We review each application before delivery.

Apply for Research

Standard

15,000 EUR net

One legal entity, one product, all its sites. Updates v1.x for 12 months. Email support.

Order Standard

Enterprise

38,000 EUR net

Everything in Standard, plus use in several products and by your affiliates, priority support and updates for 12 months.

Order Enterprise

Prices exclude VAT. Sold to companies and institutions. Read the full licence or open the datasheet. Questions? contact@xentim.com

FAQ

Questions about this dataset

Anything else? contact@xentim.com

Can I use RWF-2000 in a commercial product?

Not without approval. Its repository says the database is not allowed for commercial purposes without approval from the SMIIP Lab.

What is a hard negative in violence detection?

A hard negative is harmless behaviour that looks like a fight to a detector, such as a hug, a high five or two people dancing. Training on them helps reduce unwanted alarms.

How many people are in each event?

Between 2 and 6.

Does it include video?

No. The dataset holds skeleton data, with no RGB images or video.

Can I test it before buying?

Yes. Each dataset has a free sample and a datasheet, so you can check the format and labels on your own model first.

How do I order?

Choose a licence and send us your order. We reply with an order form and invoice, you pay by bank transfer, SEPA or card, and we send a secure download link within 3 business days of payment.

What if the data does not match the datasheet?

Tell us within 30 days of delivery. We correct the dataset or, if we cannot, refund the fee against deletion of all copies.

Test it on your model first.

The sample is free. If the data does not match its datasheet, you have 30 days to tell us.

Get the free sample