About the Project
This project relates to image classification for Mixed Martial Arts (MMA).
MMA is known as a chaotic and brutal confrontation of athletes that take advantage of numerous techniques originating from other combat sports like boxing, muay-thai, brazilian jiu-jitsu, judo, and others.
The aim is to develop a model recognizing fight scenes taking place in ‘the cage’.
Dataset contains images collected and annotated by the author, divided into:
- Training set: about 3,400 pictures augmented to 10,100 pictures
- Validation & Test set: 700 pictures each
Ultimately, a model based on DenseNet architecture was chosen.
The model obtains an accuracy level of 70%, but properly recognizes ground fight and strikes.