Vic McKracken - fcbpaint-image released at Revision 2017

Discuss anything related to the VIC
User avatar
beamrider
Vic 20 Scientist
Posts: 1447
Joined: Sun Oct 17, 2010 2:28 pm
Location: UK

Re: Vic McKracken - fcbpaint-image released at Revision 2017

Post by beamrider »

Thanks for the detailed reply.

Actually, I didn't estimate (over or under) the amount of effort involved as it's not an area I have any experience of - just posing the question :D

Just a thought - do you think the problem could perhaps benefit from an AI learning approach or genetic algorithm?
User avatar
Mike
Herr VC
Posts: 4816
Joined: Wed Dec 01, 2004 1:57 pm
Location: Munich, Germany
Occupation: electrical engineer

Re: Vic McKracken - fcbpaint-image released at Revision 2017

Post by Mike »

... or some heuristics or neural network? ;)

Main issue: there is a search space for the result that needs to be restricted somehow. In case of FCB, one can handle the picture as 48 strips of 4 pixels height with minimal coupling (over the error distribution) from one strip to the next, that can independently be optimized.

Now what those quantizers don't do is - "look" at the image and decide "take that colour/attribute" for a certain area *). Rather, they need to try out all colours that are available from the palette and take the one with the smallest error. In the case here, there are *hard* hardware restrictions that need to be modeled. Usually one takes the areas with the biggest area and thus biggest constraint, fixes the attribute data there, and then works down to the picture parts with the least constraints, the single pixels. Calculates the error for each level, hands it up to the level one up, up to the top. The attribute setting with the smallest error and the corresponding bitmap data then ends up in the result.

With FCB, the next smaller pixel areas below the 48 strips are either the 32 pixels of each foreground attribute cell, or the splits of - actually! - indefinite length, could be anything between 16 and 168 pixels. With the splits, any change in one line of the four will most probably lead to another 'optimal' distribution of splits in the other three lines - mainly because there are the foreground cells that couple those four lines. Your search space has just exploded. Unless one finds a *really* *good* heuristic, this is just a recipe to burn one's own fingers. ;)

I rather spent just 10 hours of hardware hacking to enable VFLI graphics on my VIC-20, and got nice results with photographic input which are very unlikely to be bettered by FCB, even in the absense of in-line splits. :)

But then, FCB is just another use case for graphics. The pictures done in fcbpaint thus far show that it is an excellent tool for pixelling, but there also still is some leeway for improvement. Means, the people still have to learn how to use fcbpaint. And they shouldn't restrict themselves to that single tool only: on Revision, exin confessed to me, that he did the first workstages of 'Doozy' in MINIPAINT, then converted it into FCB format (with the converter I provided here) and then made the final touches, especially the colouring, in fcbpaint.

If there is an 'agenda', that could be contributed to this release, then exactly this: there *is* a good working set of tools for the VIC-20 available to do pixel graphics. The VIC-20 is an interesting and demanding platform for pixel art. Graphicians, it's your take!

Cheers,

Michael


*) that's what our brain does, actually. There's the 'hardware' to handle patterns and colour distributions, and that does accept some deviance from the optimal result, as long as it 'looks' acceptable.
naujoks
Vic 20 Drifter
Posts: 35
Joined: Tue Aug 16, 2005 6:25 am

Re: Vic McKracken - fcbpaint-image released at Revision 2017

Post by naujoks »

50 hours! Goodness me!
Post Reply