Link to the Past level pack
 

Hey if I could dig it up and upload it, would anyone be interested in the Link to the past overworld pack I generated a while back?

Yes.

That sounds like a pretty awesome idea.

Yes!

Here you are gentlemen. I think it is mirrored and there isn't a gmap right now.. I lost it.

Not bad. Ill check them out when I get homr.

I had a type 1 and layered version I was working on, but lack graphical talent so I stopped at the lost woods. Lol, this will help though..

Mine was generated from screenshots :P It wasn't a manual effort!

I think I asked you about this a while back, but you don't have the stuff to do that anymore do you?

Nope :P Sorry.


If you want a quick rundown on how you can start to build something like I had...

Get a large screenshot of what you want to build(level size).

Using:
http://bobpowell.net/lockingbits.aspx
Scan the tiles in the tileset once and then match them up with tiles in the level.

sick

That link has some interesting information. Thanks for posting, I'll definitely take a look at this.

Attached is a GMAP file for this level set. (rename it to .gmap to use with gonstruct -- the program I tested it with).

For some reason, Jacob, the first 3-4 levels on the north-eastern side do not work (a1 - d1?), they do not draw anything for me with the given tileset. And what is drawn, it looks like the wrong images are in those locations.

Also, the ZIP file included is missing the level: a4

Thanks again.

Been reading through the article you posted on analysing image data and I've spent some time creating an application that can generate a Graal level based on a full-sized level screenshot. I've still not come up with a solution to deal with images covering tiles, light effects and level with layers however, but it's working perfectly for level images that show only tiles.

Here's an image. I've managed to generate the level on the right by taking the level image on the left and analysing it's pixel data, comparing the data with the pixel data for pics1.png. The comparison are made on a tile-by-tile basis, so every 256 pixels (or 1 tile) a comparison is made.

I'll post a build when I've figured out exactly how I'm going to deal with **** like light effects and layers.

The big question-- how long does it take? And I don't think there will be a good solution for dealing with lights or npcs in screenshots.







Seems I lost level a4 somewhere and the a1-d1 are mountain levels... they would've used a different tileset which I must not have bothered to upload at the time!

It actually takes a tremendous amount of time, but I wasn't expecting anything else. Regarding dealing with tile obscurities though, it has been suggested that I use an algorithm to check for similarities between tiles instead of directly comparing pixel data. Obviously, it's not fool proof and I doubt I'm capable of coming up with something that is, but it's a step in the right direction ;)

I think mine got down to something insane like 20 seconds for a level the more I worked with it and changed how I did things.


I am a noob but heres kind of some rambling advice as to how I was doing it and saving time.

Each scan read the r,g,b color of each tile in an X(Top left to bottom right, top right to bottom left). That is scanning 32 pixels per tiles instead of 256 (all of them)-- this increased my speed by 8 times.

I stored that in a big array from the tileset.


The scanning of the level is where time was made up:
Then when scanning the level I scanned 0,0-- looped through and checked it against the array of the tileset tiles. If a match was found-- I reorganized the tileset array of tiles to move the tile that was found to the 0 position. That way, common tiles in the level are closer to the first checked.

If a tile wasn't found-- I put it in a "not found" array. Which it'd scan before the tileset. Since there usually weren't 100s of not found tiles this stopped the thing from looping through all 4000 or so tiles each time one wasn't found.

You've suggested some nice ideas here. Yeah, the time it takes to initially scan through to find matches takes forever. Right now I'm doing something like this:

1. Add bitmap data for each tile on the tileset to a list
2. Grab the bitmap data of each tile in the level image and generate a hash for comparrison using the following method.
   
   
   HTML Code:

   ---

   private unsafe string GetPixelHash(BitmapData bd)
{
    string final = "";

    for (int y = 0; y < bd.Height; y++)
    {
        byte* row = (byte*)bd.Scan0 + (y * bd.Stride);
        for (int x = 0; x < bd.Width; x++)
        {
            final += row[x * 4].ToString();
        }
    }

    return final;
}

   ---

3. Cross compare
4. Get tileset position of tiles
5. Generate

Quite obviously a much more efficient method of comparison is possible since you've said that it's possible to cut down the generation time to around 20~ seconds. In my defence though, the code hasn't been re-written at all so you're really looking at my preliminary method at work!

Mine was a bunch of tard-code. So if you work hard at it I'm sure you can do better!

Also you could pretty easily use the streamwriter class and pump out .nw levels too!

http://www.dotnetperls.com/streamwriter