Mclaneinc Posted June 1, 2012 Share Posted June 1, 2012 That Iron Man is pure class........Remember to set to PAL Top notch... Quote Link to comment Share on other sites More sharing options...
snicklin Posted June 1, 2012 Share Posted June 1, 2012 (edited) I really enjoyed that film clip too. Well, not so much the film itself but the concept that you can almost believe that it's a film on the Atari!! OK, well here's what I've been wasting my electric on, after 119M evaluations, a scene from Alaska. output.xex Edited June 1, 2012 by snicklin 1 Quote Link to comment Share on other sites More sharing options...
emkay Posted June 1, 2012 Share Posted June 1, 2012 Just for fun, here's another video that abuses the fact that block loads are instantaneous in Altirra and AtariWin800. Brilliant! We have to wait for a version running on the real thing. Then it is brilliant. Now it's only great 1 Quote Link to comment Share on other sites More sharing options...
Xuel Posted June 1, 2012 Share Posted June 1, 2012 How long did it take to render the images? About a day on 3 cores with the case fans on full blast. Wife asked, "Why is your computer so loud now?" 200000 evaluations per frame, 2092 frames. Btw, next time try to play with higher values of /dither_val (=2 or 3) if your images get too gray. It works especially better with 'chess' dithering. Also /distance=ciede may solve the grayness problem, but the initial preprocess can take a lot of time then. I'll give dither_val a try! I really like the results of ciede but, man, it is slow. The IM clip may have been a bad choice to really show off RastaConverter since the film uses heavy filtering which squelches everything but blues and oranges, so it tends to look like B&W plus highlights of color. Nonetheless, RastaConverter is awesome. Quote Link to comment Share on other sites More sharing options...
+Stephen Posted June 1, 2012 Share Posted June 1, 2012 This didn't want to run for me in Altirra last night. Any special settings? I had disk set to burst mode, PAL, 320k RAM. Quote Link to comment Share on other sites More sharing options...
Xuel Posted June 1, 2012 Share Posted June 1, 2012 This didn't want to run for me in Altirra last night. Any special settings? I had disk set to burst mode, PAL, 320k RAM. Hmm... What symptoms did you see? The only settings that I could find that don't work are: - 16K - 5200 - NTSC Otherwise, it seems to work with any other settings. If you enable basic the bottom of the screen is garbled, but it still plays. Burst settings don't apply to the image loader -- it's always 0-cycles to load the next block. Quote Link to comment Share on other sites More sharing options...
emkay Posted June 1, 2012 Share Posted June 1, 2012 I did just drag and drop onto Atari800win and Altirra. Both works fine. Quote Link to comment Share on other sites More sharing options...
+Stephen Posted June 1, 2012 Share Posted June 1, 2012 This didn't want to run for me in Altirra last night. Any special settings? I had disk set to burst mode, PAL, 320k RAM. Hmm... What symptoms did you see? The only settings that I could find that don't work are: - 16K - 5200 - NTSC Otherwise, it seems to work with any other settings. If you enable basic the bottom of the screen is garbled, but it still plays. Burst settings don't apply to the image loader -- it's always 0-cycles to load the next block. I dragged the XEX onto Altirra, and it just stayed on the blue "bootup" screen. Quote Link to comment Share on other sites More sharing options...
Goochman Posted June 1, 2012 Share Posted June 1, 2012 You need to run it under PAL - if you have NTSC it wont start. Quote Link to comment Share on other sites More sharing options...
+Stephen Posted June 2, 2012 Share Posted June 2, 2012 Looks incredible! Do you think that 70 kB/sec is fast enough to do this? My IDE 2.0+ can manage that with screen DMA on. Quote Link to comment Share on other sites More sharing options...
Rybags Posted June 2, 2012 Share Posted June 2, 2012 Our IDE interfaces would slow down to probably under 40% their top speeds since these pics use a display kernal which means 0% CPU available for however many scanlines the pic is. On NTSC the hit would be even worse since any given display always takes a higher percentage of available time. Quote Link to comment Share on other sites More sharing options...
Xuel Posted June 2, 2012 Share Posted June 2, 2012 (edited) Looks incredible! Do you think that 70 kB/sec is fast enough to do this? My IDE 2.0+ can manage that with screen DMA on. Glad you got it working! Full screen at 25fps requires about 500kB/sec, but PBI memory mapping should be able to come to the rescue. Edited June 2, 2012 by Xuel Quote Link to comment Share on other sites More sharing options...
snicklin Posted June 2, 2012 Share Posted June 2, 2012 Just as a silly question to all, what is the highest number of evaluations that you've allowed a picture to go to? Mine is 130M but I'm sure that some people have much higher than that. Quote Link to comment Share on other sites More sharing options...
emkay Posted June 2, 2012 Share Posted June 2, 2012 Just as a silly question to all, what is the highest number of evaluations that you've allowed a picture to go to? Mine is 130M but I'm sure that some people have much higher than that. As I've mentioned before , 1000 million for some pictures. But the changes get very low there. Quote Link to comment Share on other sites More sharing options...
emkay Posted June 2, 2012 Share Posted June 2, 2012 Another question... If you have this picture.... which result would you prefer ? a) b) ? Quote Link to comment Share on other sites More sharing options...
snicklin Posted June 2, 2012 Share Posted June 2, 2012 (edited) @Emkay - (B) - It brings out the foreground better. Edited June 2, 2012 by snicklin Quote Link to comment Share on other sites More sharing options...
KubaCZ Posted June 2, 2012 Share Posted June 2, 2012 a) For me, it looks more like original one. Quote Link to comment Share on other sites More sharing options...
emkay Posted June 2, 2012 Share Posted June 2, 2012 You should watch both pictures in fullscreen. The 1st one has a screen full of dither pixels. The 2nd (b)one has a low count of dither pixels. (B) takes a side step (searching the next close colour)to use the amount of colours that are really available in the A8 palette, instead of using the dither pixels. So it leaves the colour range a bit, but the image gets clearer. Quote Link to comment Share on other sites More sharing options...
Tezz Posted June 2, 2012 Share Posted June 2, 2012 (edited) I did the SOB picture in G2F in 2004, it should be in the archive if you want to see the solution I chose. Edited June 2, 2012 by Tezz Quote Link to comment Share on other sites More sharing options...
emkay Posted June 2, 2012 Share Posted June 2, 2012 I did the SOB picture in G2F in 2004, it should be in the archive if you want to see the solution I chose. Could you post a working exe? Quote Link to comment Share on other sites More sharing options...
emkay Posted June 2, 2012 Share Posted June 2, 2012 (edited) If you try to put the picture directly to a A800 format, without dither, you get such: (Remind that has to run through the converter)... Edited June 2, 2012 by emkay Quote Link to comment Share on other sites More sharing options...
emkay Posted June 2, 2012 Share Posted June 2, 2012 (edited) As you might have recognized. Many pictures use a full screen of dither pixels, while others don't show colours where they are simply available. It would be really great, to have such a filter in the converter, using the amount of colours better for the image. Filter rules are rather simple.... 1. use the closest colour 2. check the amount of colours 3. make a crosschek with the real palette 4. exchange further colours to add details. 4a. closest colour check 4b. for missing brightness values, a step to the next available colour is the solution. Sorting rules: green is recognized brighter than brown yellow is recognized brighter than green. and so on. As you might see in the "b" picture, the trees get more visible by the green colour, while the visual depth is still there. Edited June 2, 2012 by emkay Quote Link to comment Share on other sites More sharing options...
Tezz Posted June 2, 2012 Share Posted June 2, 2012 Could you post a working exe?ok, I'll attach it here when I get home tonight. Quote Link to comment Share on other sites More sharing options...
snicklin Posted June 2, 2012 Share Posted June 2, 2012 Wouldn't it be interesting if a similar concept to this was created for music? It would probably work something like this.... You tell it how much space that you want to use, 32K for example. Then it breaks the song down into it's constituent parts (like how an mp3 encoder works with all those fourier transforms), then it creates different instruments and applies those to the song, overlaying other sounds on other channels. The instruments would then be updated according to how much space they're allowed to use. However it is encoded, it would be in a format which would fit a 'player' routine which the Atari could handle easily. It'd almost be like an AtariMP3. It would have to work iteratively like RastaConverter. To be honest, it'd be far more complex than doing it for images. It would also probably give terrible results. Though it may well be useful when given mod files from machines from a similar era, they'd also be easier to parse. Quote Link to comment Share on other sites More sharing options...
VladR Posted June 2, 2012 Share Posted June 2, 2012 6. I started recently CUDA programming so maybe in the future we will have this converter much faster 8. Multiprocessing won't increase the speed. The slowest part is execution of raster program and it can't be parallelized. If it can not be parallelized then CUDA is no option. As far as I know the vector-computing efficiency of a (CUDA enabled) graphics card is not much better then on a modern CPU. Where CUDA stands out is when you can dispatch the task to 1000 Vector ALUs. Must admit though that my CUDA experience is 3 years old. There might have benn changes. Would be interested to hear about that. However, I find it really cool that it seems to take hours on a modern PC to calculate an optimal A8 picture. I just found this thread and havent looked at the program. Do you try out every possible combination of GR15 and PMG and measure the "distance" from the original? Is there any particular reason why this couldn`t be ported to CUDA ? Are there any direct dependencies betweem consequent iterations ? Let`s say you have a card with 500 cores and you run 500 iterations. With each core working on a separate iteration, you`d get a result ~500 faster than if it was running on just on GPU core. Without actually implementing it, there`s no way to find out if the speed-up compared to CPU is going to be 0.5:1 or 100:1. How much work there is after a single iteration is done with computations ? Are there a lot of memory transfers (those are usually the ones that make the cuda speed-ups lower than expected) at the end ? 1 Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.