VideoParser

Have you thought to yourself: 'There's a lot of gameplay video out there. I (a games researcher) would sure like to get at all the info in that game in a way that is maybe not perfect but gets the job done'? Well then this is the repo for you!

What's in Here

This repo contains three python scripts to help extract information on sprite usage per frame of a video. It also contains a sprite palette and gameplay video for a game that is almost believably not a clone of Super Mario Bros. to be used in the below tutorial.

Tutorial: Introduction

In this tutorial I'll go through the following steps:

1. Where to find python 2.7 and the relevant libraries.
2. How to use the "VideoParser.py" python script to extract information from video
3. How to use the "ResizeSprites.py" python script to resize the sprite palette to better match the video.
4. How to use the "VisualizeFrames.py" python script to check the extracted information

Tutorial 1: Relevant Libraries/Downloads

Python 2.7: https://www.python.org/downloads/ Install pip: http://pip.readthedocs.org/en/stable/installing/ Install numpy: run "pip install numpy" in terminal Install pillow: run "pip install pillow" in terminal Install OpenCV: Mac: https://jjyap.wordpress.com/2014/05/24/installing-opencv-2-4-9-on-mac-osx-with-python-support/ Windows (not verified): http://docs.opencv.org/master/d5/de5/tutorial_py_setup_in_windows.html#gsc.tab=0 Download ffmpeg (video parser): https://www.ffmpeg.org/download.html

#Tutorial 2: VideoParser.py With all the relevant downloads downloaded, download the repo and cd into it.

The VideoParser.py script uses the following command:

python VideoParser.py

Note: If you are running Windows, you'll need to comment line 120 and uncomment line 121

So to start with the Moorio example let's run the following command in the terminal: python VideoParser.py Moorio.mp4 moorioSprites 1

After a few moments you'll see a new "frames" directory pop up with 28 frames (1 frame per second of the video). You'll also see a framesDescription.csv file that is basically empty. That's no good!

If you open up one of the created frames (say "image-00000001.png") you'll see that the sprites in the image are much larger than the individual sprites in "moorioSprites". This is a typical problem in parsing video, and so the next phase helps to solve that.

#Tutorial 3: ResizeSprites.py The command to run ResizeSprites.py looks like:

python ResizeSprites.py or python ResizeSprites.py

This will create a new directory called "resizedSprites" in the passed in spritesDirectory. Normally you'd want to find the resizeRatio by determining how large (in pixels) a given sprite was in comparison to how large the sprites you have are, but in this case I know it's 2.23. So go ahead and run:

python ResizeSprites.py moorioSprites 2.23

Now go ahead and rerun the VideoParser command with the new "resizedSprites" directory, with the command: python VideoParser.py Moorio.mp4 moorioSprites/resizedSprites 1

Wait a few moments then check "frameDescriptions.csv". It should have quite a bit more in it this time! Let's check exactly how well it did.

#Tutorial 4: VisualizeFrames.py

VisualizeFrames.py is a script to visualize a frameDescriptions.csv file to see how well the VideoParser.py did. The command to use it looks like:

python VisualizeFrames.py <the directory to place the visualizations

So in this case let's run the command:

python VisualizeFrames.py moorioSprites/resizedSprites frames/frameDescriptions.csv visualizedCV

You should see a new directory "visualizedCV" that contains the visualizations of frameDescriptions.csv. These are pretty good! In the last section of this ReadMe I'll point out a few possible ways to up the output, but this is about as good as you'll get.

And that's it! Following that same set of steps you should be able to get a frameDescriptions.csv file. From there you can use the sprites and their positions per frame to extract all sorts of information about a gameplay video.

#Tips and Tricks Here's a couple additional suggestions for you: -OpenCV cannot handle transparency. Your best bet is to include a background color on all sprites that either matches the background color of the game, have multiple different versions of each sprite with different background colors are limited, or create sprites with random noise as the background image.

-VideoParser.py has a number of ways to accept more or less possible sprites in a frame. The easiest way to do this is to adjust the "threshold" variable of the "FrameParser" class. The lower the value, the more (less likely) sprites will be found. However there are many more ways commented on throughout the file.