RAGING DOWNLOAD

Applying data science to fighting games.

Each action a player performs in a match is recorded into a dataset which can be analyzed to find trends in their play.

Objectives

1. Use data analysis to achieve deeper gameplay analysis.
2. Provide conveinent gameplay analysis for new and casual players

Read the article for more information

Daigo vs. Tokido Demo (SFV)

Proof of concept that uses manually collected data from a video to analyze a match and prove how useful this data can be in fighting games.

Sprite Detection

Using object detection to collect data from videos of fighting game matches without manually collecting data.

Click for the full video

HOW TO USE

Currently the sprite detection is not extremely user friendly, but if you want to try the sprite detection on your own videos try following these steps. (Requires Python and yolov5 dependencies to be installed.)

1. Download the Sprite Detection folder
2. Download the yolov5 repository and place it in the Sprite Detection folder
3. Run detect.bat (Edit the filepath for the source to use your own video and not the sample)
4. Object detected images or videos will be saved to yolov5/runs/detect/exp

Utilizes sprites from:

https://www.justnopoint.com/zweifuss/

https://srk.shib.live/w/Street_Fighter_3:_3rd_Strike

http://ensabahnur.free.fr/BastonNew/index.php

Data generated by YARDS.

---

WORK IN PROGRESS

• Write script that translates detection output into useful data
• Recreate Ken model with more classes and better training
• Attempt model with two or more characters
• Apply sprite detection to Under Night
• Clean up detection data
• Create user-friendly GUI for detection