New "higheSTcolor" mode better than Photochrome & Spectrum

Discuss your favourite artists, styles, art packages and much more in this section

Moderator: Moderator Team

User avatar
DarkLord
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 4178
Joined: Mon Aug 16, 2004 12:06 pm
Location: Prestonsburg, KY - USA
Contact:

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby DarkLord » Mon Nov 21, 2011 1:02 pm

Quick question, does everything involved with this format have to be at
a stock 8mhz or the pictures will be garbled?

Just wondering...and if so...is there any way to ever get it to work with
accelerated Atari's?

Only asking this because I seem to remember not being able to get
Photochrome pictures to work right on my AdSpeed equipped Mega
ST until I kicked it back down from 16mhz to 8...

Thanks!
Welcome To DarkForce! http://www.darkforce.org "The Fuji Lives.!"
Atari SW/HW based BBS - Telnet:darkforce-bbs.dyndns.org 520

User avatar
Cyprian
Atari God
Atari God
Posts: 1402
Joined: Fri Oct 04, 2002 11:23 am
Location: Warsaw, Poland

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby Cyprian » Tue Nov 22, 2011 11:28 am

Zorro 2 wrote:Bravo Mister Cyg :cheers:

Here the final result : Street Art (http://pouet.net/prod.php?which=58056)


yeah!! that's impressive!! my congrats Cyg!! We're waiting for next prod!!

DarkLord wrote:Quick question, does everything involved with this format have to be at
a stock 8mhz or the pictures will be garbled?

works fine on 8mHz, you can see live version in this prod: http://pouet.net/prod.php?which=58056
Jaugar / TT030 / Mega STe / 800 XL / 1040 STe / Falcon030 / 65 XE / 520 STm / SM124 / SC1435
SDrive / PAK68/3 / CosmosEx / SatanDisk / UltraSatan / USB Floppy Drive Emulator / Eiffel / SIO2PC / Crazy Dots / PAM Net
Hatari / Aranym / Steem / Saint
http://260ste.appspot.com/

Cyg
Atariator
Atariator
Posts: 25
Joined: Fri Apr 29, 2011 2:03 pm

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby Cyg » Wed Nov 23, 2011 12:42 am

Thanks :)

Already thinking & working on the next prod with new or more advanced FX and a better design.
Zorro, this was not a"final" result but something like a reSTart :)

DarkLord, as Cyprian_K said, yes you need to be at 8Mhz to run it (ST/STE/MegaST) as most of the effects are very precisely synchronised with the 68000 instructions timings.

User avatar
DarkLord
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 4178
Joined: Mon Aug 16, 2004 12:06 pm
Location: Prestonsburg, KY - USA
Contact:

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby DarkLord » Wed Nov 23, 2011 2:34 am

Cyg wrote:DarkLord, as Cyprian_K said, yes you need to be at 8Mhz to run it (ST/STE/MegaST) as most of the effects are very precisely synchronised with the 68000 instructions timings.


Gotcha. I kinda figured that. Its not a problem - most accelerators allow you to disable them, or down shift their speed on the fly. I was just curious.

Thanks for such a nice effort. I personally love demo's and "eye candy". I can sit in front of my monitor and zen out, watching them. :)
Welcome To DarkForce! http://www.darkforce.org "The Fuji Lives.!"
Atari SW/HW based BBS - Telnet:darkforce-bbs.dyndns.org 520

User avatar
dml
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 3472
Joined: Sat Jun 30, 2012 9:33 am

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby dml » Mon Aug 27, 2012 8:25 pm

Interesting thread.

It seems in this post http://www.atari-forum.com/viewtopic.php?f=81&t=21148&hilit=photochrome#p188331 your test is rendered with dithering, but the other formats are rendered out without dithering? Why is that?

It is not too difficult to get more colours per line but the problem is packing in the density of changes within the image itself. Ideally loading the colours in the border and displaying them within the visible line. It is however still useful to have the extra colours because it means less borrowing of colours carrying the previous scanline. I was never sure which would win but it seems like more colours may be better even if there is lower density in the image itself.

The quantization algorithm carries a lot of the work though, and this is where I'm concentrating again.

Anyway nice work. I'm improving PCS again so we'll see what can be done.

kristjanga
Captain Atari
Captain Atari
Posts: 400
Joined: Sat Jul 25, 2009 3:35 pm

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby kristjanga » Wed Aug 29, 2012 2:21 pm

perhabs a stupid question :?
but can a sampled sound be played along with animation?

User avatar
Nyh
Atari God
Atari God
Posts: 1496
Joined: Tue Oct 12, 2004 2:25 pm
Location: Netherlands

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby Nyh » Wed Aug 29, 2012 3:24 pm

kristjanga wrote:perhabs a stupid question :?
but can a sampled sound be played along with animation?

Yes, it can be done. See for example "The Snowman" animation.

Hans Wessels

User avatar
nativ
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 4087
Joined: Mon Jul 30, 2007 10:26 am
Location: South West, UK

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby nativ » Wed Aug 29, 2012 3:25 pm

kristjanga wrote:perhabs a stupid question :?
but can a sampled sound be played along with animation?


There was an old demo with Spectrum 512 ( GFA raytrace ) animation and a looped sample

viewtopic.php?f=1&t=21459&hilit=+genesis
Atari STFM 512 / STe 4MB / Mega ST+DSP / Falcon 4MB 16Mhz 68882 - DVD/CDRW/ZIP/DAT - FDI / Jaguar / Lynx 1&2 / 7800 / 2600 / XE 130+SD Card // Sega Dreamcast / Mega2+CD2 // Apple G4

http://soundcloud.com/nativ ~ http://soundcloud.com/nativ-1 ~ http://soundcloud.com/knot_music
http://soundcloud.com/push-sounds ~ http://soundcloud.com/push-records

User avatar
dml
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 3472
Joined: Sat Jun 30, 2012 9:33 am

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby dml » Fri Aug 31, 2012 11:15 pm

Here's the same image rendered out with PhotoChrome v5, using the original limitation of 48 colour changes per scan.

Image

This is without Floyd diffusion (just stochastic dithering for lost bitdepth and cross-dithering for field interlace) but using my new colour reduction algorithm, which is a significant improvement over the old one. Note significant improvements to colour reproduction on areas with gradients and colours with low population count. The new system uses visual perception rules to manage error calculation, among other things.

Obviously 48 colours per line is limiting but I have been mainly concentrating on good colour reduction and flexibility in the convertor. It will be possible to apply the 55-colours per line rule (or any other rule) to generate images for any desired display routine.

Cyg
Atariator
Atariator
Posts: 25
Joined: Fri Apr 29, 2011 2:03 pm

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby Cyg » Tue Oct 02, 2012 10:50 pm

I also improved Hicolor since my first post more than 1 year ago, in fact I found a bug on the quality estimator that lead to choose the 2nd best result at each step, leading to an acceptable result but not the best possible...

Here are the new results, I've doubled the size to see it closer.

Image
Previous Hicolor results with the quality bug, 1pix, 1pal, no interleaced, 12 bits (4096 colors) with a floyd stendberg dithering

Image
Hicolor 1pix, 1pal, no interleaced, 12 bits (4096 colors) with a floyd stendberg dithering - 762 distinct colors

Image
Hicolor 1pix, 1pal, horizontal line interleaced (little TV effect on emulators), 15 bits, no dithering - 3191 distinct colors

Image
Hicolor 2pix, 2pal, interleaced per pixel (alternatively lighter or darker), 15 bits, no dithering - 3800 distinct colors - 780 & 800 colors on individual swaping pictures

A last attemp gave better results :
Image
Hicolor 2pix, 2pal, interleaced 15 bits, no dithering - 4200 distinct colors - first picture is the 1pix + 1pal result above (762 distinct colors), 2nd picture is the opposite errors of the first compared to the original ( perfect - (1st pic - perfect) ) - 960 colors. In average the 2nd pic decreases the gap error but with additional noise from 1st pic errors

Image
Photochrome 5 as previously rendered by dml (resized x2)
Last edited by Cyg on Wed Oct 03, 2012 11:21 pm, edited 3 times in total.

FedePede04
Atari Super Hero
Atari Super Hero
Posts: 951
Joined: Fri Feb 04, 2011 12:14 am
Location: Denmark
Contact:

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby FedePede04 » Wed Oct 03, 2012 9:43 am

what a nice quality, your algorithm produce. :thumbs:
Atari will rule the world, long after man has disappeared

sometime my English is a little weird, Google translate is my best friend :)

User avatar
dml
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 3472
Joined: Sat Jun 30, 2012 9:33 am

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby dml » Tue Oct 16, 2012 12:23 pm

Beautiful - you are getting really impressive results from your convertor! :D

When I got back from holiday I quickly reworked most of PCS to remove all of the approximations I kept to make the simpler conversion modes run quickly on 8mhz 68000. The better conversion algorithms are very costly so keeping those compromises was just limiting progress with image fidelity.

So after these changes I think I am rapidly converging on the limits for image quality while remaining compatible with the old PCS display format (48cc per line, uneven cc distribution). There are a few more things to try with the m3 algorithm but it's probably getting close to the limit now without changes to display code.

This was the last result I got. Seems to be better preservation of detail in most areas and lower overall bias (bias = 0.0915% colour offset from source image). No dithering is used in this test.

scaled up:
Image

320x200
Image

Cyg
Atariator
Atariator
Posts: 25
Joined: Fri Apr 29, 2011 2:03 pm

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby Cyg » Thu Nov 01, 2012 8:44 am

Hi dml,

I think there is an issue somewhere in your 2 pics, the 320x200 is made up of 4087 different colors and not the exact aspect ratio as the original (something like one more line and a blurry result, resize issue from 199 to 200 pixels in heigth ?) and the resized 640x400 is made up of more than 140000 colors and blurry I believe because of a non preserved resize process (bilinear or something like that instead of nearest color) which is not possible :-)

I can see more than "16 shades of RVB", like in this hexa sequence : 99 88 91 80 77 6F 77 , it rather looks like 32 shades, then a 2 swapping pics result.
You choosed to multiply by 17 the Atari shades into the 24b color space, as Hatari choosed. I think I'd better do that also to simulate the final result, instead I multiply by 16 but the 24b result will never display a full white but a very light grey (same for any RVB maximum value).

What is the formula of your bias % so I can use exactly the same, I already use an absolut manhattan distance, a relative manhattan distance to check if the colors are well balanced (not too dark or light), distance from a perfect 24b to 12b conversion (as if we had no display limits with 4096 colors always available per pixel) , and square distance on the RVB space and the RVB with weight stronger on V than on R and B (to simulate our eyes preferences), this last error distance is the one I use during my optimisation process.

Cyg

User avatar
dml
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 3472
Joined: Sat Jun 30, 2012 9:33 am

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby dml » Thu Nov 01, 2012 10:23 am

Hi Cyg,

Cyg wrote:I think there is an issue somewhere in your 2 pics, the 320x200 is made up of 4087 different colors and not the exact aspect ratio as the original (something like one more line and a blurry result, resize issue from 199 to 200 pixels in heigth ?)


The image was exported as a PNG via FreeImage from the PCS image fields, but there might be an accidental resize happening that I didn't notice. Although if it has added 'extra colours' it will be a product of a bilinear filter squashing/stretching it by one line or something like that. I'll repeat the test and see what happened.

Cyg wrote: and the resized 640x400 is made up of more than 140000 colors and blurry I believe because of a non preserved resize process (bilinear or something like that instead of nearest color) which is not possible :-)


That was me. I accidentally resized in PSP with bilinear on and didn't notice until I looked later. It's just the other one doubled in size so it can be corrected. It's probably better that I check the 320x200 image though for the other issue.

Cyg wrote:I can see more than "16 shades of RVB", like in this hexa sequence : 99 88 91 80 77 6F 77 , it rather looks like 32 shades, then a 2 swapping pics result.


It's 16 shades interlaced which produces 30 distinct shades (it was a 2-field image without dithering).

Cyg wrote:You choosed to multiply by 17 the Atari shades into the 24b color space, as Hatari choosed. I think I'd better do that also to simulate the final result, instead I multiply by 16 but the 24b result will never display a full white but a very light grey (same for any RVB maximum value).


Yep originally I was just chopping into 0-240 (e.g. 8bit >> 4) like the 1992 version of PCS written in 68k, but I changed things a couple of times since. It now scales 0-30 into 0-255 to use the full range - although it makes diagnostics a bit difficult because of the uneven quantization between those two spaces. I therefore perform diagnostics in both spaces now (source 8bit, and native-virtual / 15 or 30 shades) since the distance to both can vary on any image. e.g. you can get a perfect result in virtual depth but a large error in 8bit, or you can get closer to 8bit than virtual in squared error space because of the nonlinear summing.... there's almost no end to the fun :)

Cyg wrote:What is the formula of your bias % so I can use exactly the same, I already use an absolut manhattan distance, a relative manhattan distance to check if the colors are well balanced (not too dark or light), distance from a perfect 24b to 12b conversion (as if we had no display limits with 4096 colors always available per pixel) , and square distance on the RVB space and the RVB with weight stronger on V than on R and B (to simulate our eyes preferences), this last error distance is the one I use during my optimisation process.
Cyg


IIRC I used different measurements for each diagnostic pass (the diagnostic modules are kindof plugins that run separately). The 'bias' value is supposed to show you if your image is offset (lighter or darker) so it is the average error. i.e. if you compare a black and white image the bias would be +100%, and if you invert the images you should get -100%. So the aim is to get a bias less than one intensity unit of the source (1/255) or better. Bias doesn't tell you about peak error, which is where Euclidian and Manhattan distances get used. In fact bias, assuming it is not falsely compensated, becomes 'noise' as other types of error dominate so providing it's kept small the absolute value doesn't really tell you much. But if bias is consistently a few intensity notches above or below zero it probably indicates a real offset.

There are 3 other diagnostics - one measures summed Euclidean distance, another measures summed Manhattan, the last measures the sum of the squares of all the R,G,B errors (SquERR) and is the simplest figure to check when refining faults, albeit imperfect. Each pass runs in the 2 colour spaces (8bit and native-virtual).

I also emit some diagnostic images, one per diagnostic pass. The 'bias' image uses green for pixels 'too low', blue for 'too high' and white/yellow for epsilon and perfect match. The aim of course is to have equal blue and green, it's just a visual check for making sure gradient testcards convert properly with error distributed both sides of a quantization step - the bias% is more accurate.

The diagnostic code is a separate source module so the calcs are seen there, it's all open and repeatable, if you excuse my loose fingers in PSP :angel:

I made quite a few changes since, including fixing a horrible omission in -m6 (it wasn't updating the error on a bin after editing it) so I kept the old -m3 for speed and added the 'fixed' one as -m6. I also added a new algorithm and did a lot of work on dithering.

Most of the problems I have now are due to being stuck with a 1990-era display routine and the difficulty of solving an image 'whole' instead of a scan at a time, which I think everyone else is probably using now, with extra colours (pre)loaded per line. So I have hit an image quality barrier with that, but I'm quite happy with the other bits now. When I return to it I will look at the display upgrades.

User avatar
dml
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 3472
Joined: Sat Jun 30, 2012 9:33 am

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby dml » Fri Nov 02, 2012 9:56 am

Cyg wrote:Hi dml,
I think there is an issue somewhere in your 2 pics, the 320x200 is made up of 4087 different colors and not the exact aspect ratio as the original (something like one more line and a blurry result, resize issue from 199 to 200 pixels in heigth ?)


I checked quickly and I don't think there is a problem with the output, and it should only rescale input if the image is bigger than 320x200. TBH I'm not sure which source image I used to generate it - it was either the original (high res) one, or the pre-sized 320x200 one. If I used the high res one then the rescaling may have made the resulting image non-comparable with the other tests I suppose (if you try to subtract them for example).

It was a dual-field image, 31 shades (0-30 inclusive) so 4087 distinct colours should be fine.

The doubled-up image though, was bilinear stretched in PSP by hand so best ignore that one!

Oh btw - another reason to be wary of small bias% measurements on random images - it's possible (but rare) for a specific colour to dominate an image, and that colour will always 'round' in one direction (assuming error diffusion is disabled). So this can throw bias off even if conversion is otherwise ideal. It is why bias will also drop to background noise level for most images - some small random looking number. It's only really interesting if it is consistently off by some reliable, significant constant with a range of images.

I suppose bias% might also be a good way to measure effectiveness of error diffusion if both are measured in linear space, since the two things should mostly cancel out! Possibly a good cross-check for both processes?

Also, the 'native virtual' space I use for quantization is as follows:

ST/single-field: 0-7 (8 levels, 512 cols)
ST/dual-field: 0-14 (15 levels, 3375 cols)
STE/single-field: 0-15 (16 levels, 4096 cols)
STE/dual-field: 0-30 (31 levels, 29791 cols)

These spaces affect the colour rounding/quantization steps in all processes. Input dithering magnitude is also relative to these spaces (e.g. input halftone-dither mode needs to know the output range etc.). It's quite difficult to get decent, reliable diagnostics without quantization in the appropriate output space for each mode.

One of the things i built into PCS510 (but did not yet document) is built-in testcards which render source images perfectly in the various colour spaces (8bit + native-virtual), and I used these to test the diagnostics, since bias% and error% should be zero when the input is pre-quantized and there are fewer colours used than the colour changer is capable of (i.e. no remap errors).

User avatar
nativ
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 4087
Joined: Mon Jul 30, 2007 10:26 am
Location: South West, UK

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby nativ » Fri Nov 02, 2012 1:17 pm

What is the highest colour Medium Res Mode?
Atari STFM 512 / STe 4MB / Mega ST+DSP / Falcon 4MB 16Mhz 68882 - DVD/CDRW/ZIP/DAT - FDI / Jaguar / Lynx 1&2 / 7800 / 2600 / XE 130+SD Card // Sega Dreamcast / Mega2+CD2 // Apple G4

http://soundcloud.com/nativ ~ http://soundcloud.com/nativ-1 ~ http://soundcloud.com/knot_music
http://soundcloud.com/push-sounds ~ http://soundcloud.com/push-records

User avatar
Scarlettkitten
Captain Atari
Captain Atari
Posts: 259
Joined: Thu Mar 19, 2009 11:42 am
Location: Northamptonshire, UK

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby Scarlettkitten » Fri Nov 02, 2012 8:53 pm

nativ wrote:What is the highest colour Medium Res Mode?


found this from here
http://pouet.net/prod.php?which=22552

Image
Music gear
Falcon 030 14MB, Cubase Audio, Soundpool FA8,FDI, MAudio 88 keystation, Roland S750, Yamaha A5000, Roland JV1080, Yamaha MG10, 1040STE, ZX Spectrum 128k.

Cyg
Atariator
Atariator
Posts: 25
Joined: Fri Apr 29, 2011 2:03 pm

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby Cyg » Fri Nov 02, 2012 9:26 pm

I think it can be more than this, like here : http://www.pouet.net/prod.php?which=12237 HighResMode by Paradox on STE, 640x400 images with up to 14,000 colours on STE machines using a blend of techniques : midres, palette/scanline interlace and blitter palette-shiftings
Image

I will try one day, it could be interesting in some case even if I believe that it's the number of colors which is more important than the number of pixels.

Cyg

User avatar
nativ
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 4087
Joined: Mon Jul 30, 2007 10:26 am
Location: South West, UK

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby nativ » Fri Nov 02, 2012 10:46 pm

Hadn't seen the tdf version, thats really interesting to see the 640*400 used with 16 colours! wonder how useable it is.

not sure if I saw the paradox hicolour before I have a feeling the flicker was too much?

Is obligatory to boost the screen res as well?
Atari STFM 512 / STe 4MB / Mega ST+DSP / Falcon 4MB 16Mhz 68882 - DVD/CDRW/ZIP/DAT - FDI / Jaguar / Lynx 1&2 / 7800 / 2600 / XE 130+SD Card // Sega Dreamcast / Mega2+CD2 // Apple G4

http://soundcloud.com/nativ ~ http://soundcloud.com/nativ-1 ~ http://soundcloud.com/knot_music
http://soundcloud.com/push-sounds ~ http://soundcloud.com/push-records

User avatar
nativ
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 4087
Joined: Mon Jul 30, 2007 10:26 am
Location: South West, UK

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby nativ » Fri Nov 02, 2012 11:02 pm

Cyg wrote:]

I will try one day, it could be interesting in some case even if I believe that it's the number of colors which is more important than the number of pixels.

Cyg



I wonder if a letterbox mode 16:9 could be made for ppera's video streamer?
Atari STFM 512 / STe 4MB / Mega ST+DSP / Falcon 4MB 16Mhz 68882 - DVD/CDRW/ZIP/DAT - FDI / Jaguar / Lynx 1&2 / 7800 / 2600 / XE 130+SD Card // Sega Dreamcast / Mega2+CD2 // Apple G4

http://soundcloud.com/nativ ~ http://soundcloud.com/nativ-1 ~ http://soundcloud.com/knot_music
http://soundcloud.com/push-sounds ~ http://soundcloud.com/push-records

Cyg
Atariator
Atariator
Posts: 25
Joined: Fri Apr 29, 2011 2:03 pm

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby Cyg » Fri Nov 02, 2012 11:44 pm

nativ wrote:I wonder if a letterbox mode 16:9 could be made for ppera's video streamer?


We will give it a try, surely with less than 200 pixels in height because of the bandwith limits and CPU to manage it for the bitmap loading/displaying, and 400 or up to 416 pixels in width, making it more like a 2.35:1 mode (as the super widescreen mode).

Cyg

AtariZoll
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 2796
Joined: Mon Feb 20, 2012 4:42 pm
Contact:

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby AtariZoll » Tue Apr 30, 2013 11:39 am

Max res. with 25 fps and 48 colors/line : 416x228 . It is aprox 16:9 .

http://atari.8bitchip.info/movpst.php
Negative feedback has usually positive effect.

Sixxorius
Atarian
Atarian
Posts: 1
Joined: Wed May 22, 2013 5:36 am

Re: New "higheSTcolor" mode better than Photochrome & Spectr

Postby Sixxorius » Wed May 22, 2013 5:40 am

Looks superb! :D


Social Media

     

Return to “Discussion”

Who is online

Users browsing this forum: No registered users and 1 guest