Hi,
I wonder if it would be possible to implement a raster -> vector graphics converter, in order to improve the looks/resolution. In special for the oldest games.
Could an integrated version of Autotrace do the job?
Wish: Vector graphics
Moderator: ScummVM Team
highly unlikely, also note that most games don't have simple 1-color gfx (DOTT and Samnmax might seem that way but there are lots of slightly off-colored pixels everywhere)
of course you could do that manually and offer those additional gfx for download - and of course run into lots of legal problems with the copyright holders
of course you could do that manually and offer those additional gfx for download - and of course run into lots of legal problems with the copyright holders
-
eriktorbjorn
- ScummVM Developer
- Posts: 3561
- Joined: Mon Oct 31, 2005 7:39 am
Re: Wish: Vector graphics
Is that really practical? For one thing, it's a rather slow process and when I tried it with an image from Loom, with default parameters, it looked like this. Telling it to despeckle the image made for some improvements, but I'm still not convinced.Micru wrote:Hi,
I wonder if it would be possible to implement a raster -> vector graphics converter, in order to improve the looks/resolution. In special for the oldest games.
Could an integrated version of Autotrace do the job?
Maybe someone more familiar with autotrace could adjust the parameters to come up with something better, but would it always be better, or just for this particular image?
Last edited by eriktorbjorn on Mon Nov 13, 2006 10:04 am, edited 1 time in total.
even better than libcaca.I don't believe using a general purpose vectorizer like autotrace has any chance of producing sensible output, ever, for raster based hand tuned (and dithered) graphics as used in particular in the older SCUMM games.
The process of converting raster to vector data is inherently hard. You simply can't recover information from a picture that isn't present in it anymore, due to low resolution. Hence a vectorizer has to "guess" a lot. And that requires fine tuning. The graphics in Loom (EGA) have very different properties from those found in Maniac Mansion v1/v2, in Loom VGA, or in Full Throttle, for example (this is also the reason why, at least IMO, for each of these games different scalers look "best"; there is no single scaler that produces optimal resoluts for all and each game supported by ScummVM. IMHO.). Therefore, to have even a remote chance of ever attaining useful output, one would have to add/apply lots of fine tuned vectorizer settings, different ones for each game (and maybe each scene). Which kind of is contrary to the purpose of a *general purpose* vectorizer...
In a sense, some of our scalers already try do some kind of "vectorization", in so far as they try to detect edges (at least locally) -- in particular, the hq2x and advmame2x scalers follow that approach. Of course both work with relatively tiny peepholes (they only consider a tiny neighborhood of each pixel they are scaling).
All in all, I do not believe that doing on the fly conversion from raster to vector data inside of ScummVM would do any good; even if we were able to deal with the performance issues, the fundamental issues with this approach render it impractical (where is the point in "high quality vector graphics" if they end up looking worse than the original raster graphics?).
The process of converting raster to vector data is inherently hard. You simply can't recover information from a picture that isn't present in it anymore, due to low resolution. Hence a vectorizer has to "guess" a lot. And that requires fine tuning. The graphics in Loom (EGA) have very different properties from those found in Maniac Mansion v1/v2, in Loom VGA, or in Full Throttle, for example (this is also the reason why, at least IMO, for each of these games different scalers look "best"; there is no single scaler that produces optimal resoluts for all and each game supported by ScummVM. IMHO.). Therefore, to have even a remote chance of ever attaining useful output, one would have to add/apply lots of fine tuned vectorizer settings, different ones for each game (and maybe each scene). Which kind of is contrary to the purpose of a *general purpose* vectorizer...
In a sense, some of our scalers already try do some kind of "vectorization", in so far as they try to detect edges (at least locally) -- in particular, the hq2x and advmame2x scalers follow that approach. Of course both work with relatively tiny peepholes (they only consider a tiny neighborhood of each pixel they are scaling).
All in all, I do not believe that doing on the fly conversion from raster to vector data inside of ScummVM would do any good; even if we were able to deal with the performance issues, the fundamental issues with this approach render it impractical (where is the point in "high quality vector graphics" if they end up looking worse than the original raster graphics?).
-
eriktorbjorn
- ScummVM Developer
- Posts: 3561
- Joined: Mon Oct 31, 2005 7:39 am
Sheesh, I was so trying to be discouraging...glokidd wrote:Wow those are some crazy loom pics
The close-ups are usually more well-defined than the normal background graphics, so the tracer would have an easier time with them, I guess. But the subtitles are problematic, as can be seen in the following images: Hetchel, Atropos, and Hetchel as a cygnet.
I'm now more convinced than ever that autotrace is useless for ScummVM.
I really hope it doesn't take that long. Actually, it isn't needed so much, just a bit more accuracy and for me it would suffice.
I don't expect on the fly conversion, just a (better) conversion to vector, and then again another conversion to raster at a higher resolution. The engine could use this new sprites instead of the old ones. You don't need so much power to do this, and if someone tried to play Loom in a 30" screen, I'm sure it would look better.
I don't expect on the fly conversion, just a (better) conversion to vector, and then again another conversion to raster at a higher resolution. The engine could use this new sprites instead of the old ones. You don't need so much power to do this, and if someone tried to play Loom in a 30" screen, I'm sure it would look better.
Nope. The point I was trying to convey (but apparently failed to) is that you simply can't "just improve the raster-to-vector conversion" a bit. It's a very very difficult and tough problem. Yeah, there are very good vectorizers our there -- but they all work with high resolution input. I.e. 300dpi upwards, better a couple 1000 dpi. Those screen shots, on the other hand, not only have very low resolutions (72dpi and less), but they also suffer from the problem that they use screen raster techniques like dithering and other things to produce certain visual effects. It is extremly difficult to deal with any of those. And even more difficult to do so in an automated fashion -- i.e. without having to readjust the settings for each scene from scratch.
It's really a pipe dream. You have better chance of convincing an artist to go over all graphics in those games and redo them, I'd say.
It's really a pipe dream. You have better chance of convincing an artist to go over all graphics in those games and redo them, I'd say.
Last month i put that sugestion in this topic.fingolfin wrote: It's really a pipe dream. You have better chance of convincing an artist to go over all graphics in those games and redo them, I'd say.
This could be a nice feature because it could be possible to:
Play the original Maniac Mansion in ScummVM with improved graphics (maybe from MMD). Maybe this could stop people asking ScummVM team to support MMD. "Fix" some EGA sprites left in Indy 3 FM Towns (that were ok in the VGA PC version).That could make (Lucas Arts) lawyers happyfac wrote:Last month i put that sugestion in this topic.fingolfin wrote: It's really a pipe dream. You have better chance of convincing an artist to go over all graphics in those games and redo them, I'd say.
This could be a nice feature because it could be possible to:
