inasavesalot

...name is still work in progress.

Basically, Ina saves quite a lot and it was most notable during her Breath of the Wild playthough. So naturally, given the EU timezone, I was watching Ina at 5 AM and the sudden motivation to do another project struck me.

A year ago? Two years ago? Dunno, time flies by quick, I made inadoki, which monitored Ina's heart rate monitor during her Horror Week. That project was written in the same way like this one, at 5 AM with a ton of caffeine coursing through my blood veins. However, this project will differ a bit from the other one, most notably by using Python instead of Bash-ing my way through it.

It is not as polished as inadoki, since I have made this in a spur of a moment and in a spent only a few hours on it.

Usage

This project is specifically tailored to detect the BoTW saving dialog and it is quite simple. To use this, you basically run:

$ python3 __main__.py -h
usage: __main__.py [-h] -g GEOMETRY -i INPUT [-ss START] [-n THREADS] [output]

positional arguments:
  output                Path to output detected timestamps. Defaults to stdout.

options:
  -h, --help            show this help message and exit
  -g GEOMETRY, --geometry GEOMETRY
                        Geometry in pixels where the dialog should appear (WxH+X+Y or "pick")
  -i INPUT, --input INPUT
                        Input video file or a still frame
  -ss START, --start START
                        Start from this frame
  -n THREADS, --threads THREADS
                        How many CPU threads to use. (Defaults to all)

Finding out the geometry

This script works by checking every frame for the rectangle defined by the geometry and if it is filled completely white. To find this geometry, you can use the -g pick, which will open the video for you and allow you to seek through the video. Dragging your cursor will draw a rectangle in the window and print out the coordinates into your standard output. To close this window either CTRL+C out of there or press q. Pressing space will also pause the playback and left/right arrow keys move you frame by frame.

Running the script

To run the script, just plug in the values as you need:

$ python3 __main__.py -g 247x36+515+412 -i o7BSE-74u8U_720.webm o7BSE-74u8U.csv -n 4

Once you are done, the output will be CSV formatted with a header, where the first column are the frame position and the second are the "amount of brightness", which should be between 250 and 255. Given that this box is also used for other UI elements, you might get some false positives such are attempting to load file etc. Also very bright areas might trigger itself as well.

You will most likely need to filter through the values to remove neighboring detected frames.

Source collection

The data I collected used 720p30, given that the part I am monitoring should be fairly distinguishable even under lower bitrate and resolution. To do this, I used yt-dlp with as following:

$ yt-dlp -S '+size' -f 'bestvideo[height=720]' -o '%(id)s.%(ext)s' <video_url>

After that, just plug it into the program itself and carry on.

License