Research  

 

Martial Arts Tricking Dataset (MATDAT) 

 

Description

The MATDAT dataset is composed by more than 90K labeled images corresponding to 3 user-generated video sequences that represent some of the most common scenarios found in martial arts tricking.

The main characteristics of this dataset are the following:

• Video sequences have been obtained using a static recording camera.
• Video sequences have been acquired by an iPhone 6S, sharing the following properties:

o Spatial resolution: 1920x1080.
o Frame rate: 29.97 fps.

• Video sequences show different background and foreground actions.

matdat1

matdat5 

 matdat2

For questions about this dataset, please contact Marcos de Rodrigo at This email address is being protected from spambots. You need JavaScript enabled to view it..

 

Ground truth description

For each video sequence in the dataset, a .CSV file with its ground truth is included in the dataset with the following content:

  • Event type:
    • Annotations include 3 different event categories: highlight (HL), not a highlight (NHL), and uncertainty (UN).
      • HL events include those frames in which a player is performing. Only HL events of a minimum duration of 1 second (i.e., 30 frames) have been annotated, since highlights of shorter duration can hardly be considered as such.
      • NHL events comprise the frames in between highlight events, where no player is performing.
      • The existence of UN as a ground truth event is due to the nature of the sport, which makes it challenging to determine the exact frames where a highlight event begins or ends. UN is used before and after every HL event as a margin of error in the manual labelling.
    • First frame:
      • First frame of a certain event.
    • Last frame:
      • Last frame of a certain event.
    • Num. frames:
      • Number of frames of a certain event.

matdat3

The table below summarizes the ground truth events found in the 3 video sequences of the dataset. 

matdat4

Download

Click here to download the MATDAT dataset (video sequences and ground truth).

 

Citation

Rodrigo, M., Cuevas, C., Berjón, D. et al. Automatic highlight detection in videos of martial arts tricking. Multimed Tools Appl 83, 17109–17133 (2024). https://doi.org/10.1007/s11042-023-16003-7.