VGMaps
General Boards => Mapping Tips/Guides => Topic started by: NeatNit on July 23, 2020, 03:38:41 pm
-
I'm playing a procedurally-generated 2D sidescrolling game, so the map is different every time.
It's definitely within the abilities of modern computers and algorithms to make a program that receives just a continuous video of the game being played, and creates a map of just the things that stay static. Even if it doesn't get perfect results, it would be super useful for one-off player usage (like procedurally generated games) and would help mappers get a good first draft for their maps.
Does such a program exist? My googling hasn't found any, but I figured if anyone would know, it's this community.
Thanks!
-
Not sure if this is what you want, but have you tried Maxim's screenshot autostitcher?
https://www.vgmaps.com/forums/index.php?topic=516.0
https://www.youtube.com/watch?v=aGR5PbvFY10&list=PL56690DF60ECAD041&index=2
-
Thank you, that's incredible. Unfortunately I don't think it's quite good enough for my needs based on the description (I didn't test it), because:
- it doesn't work on a screen capture, i.e. it requires a pre-recorded video or screenshots as input.
- It's too slow, so even if it accepted a screen capture it wouldn't be able to keep up.
- It probably won't handle variations in hue and brightness in the background - not important for NES games, but modern games make heavy use of full-screen effects (e.g. Noita (https://noitagame.com/), which is what prompted me to want this).
If you think I'm being picky, you're right, I am. The imaginary program I had in mind would work in real-time, making a map for you as you play the game. And I really think it's achievable, with the right algorithms.
I'm really intrigued by this idea now. So much so that I might pursue making it myself in the future! If I ever do, I'll be sure to post on this forum. But don't hold your breath, we're talking at least a couple of years in the future. I'm currently studying for a degree in computer engineering, and I might take some courses on image processing to learn about exactly the algorithms I'd need. Maybe.
Anyway, thank you for the link, it's exactly what I meant and Maxim sounds like an awesome guy. I'll look into using his program!
-
It sounds like what you're looking for might exist in the near future but would most likely requires some form of deep learning in order to compensate for those changes in hue and brightness. I have no idea if useful results could be achieved in real-time though.
-
You can try a program for taking screenshots, that auto-names and saves them, then stitch together afterwards