Movie Cart

[+Andrew Davie]

I'm now up-and-running with the TouchDesigner and moviecart software on my MacBook Pro (2013)

It's an interesting environment. Strange, and slightly odd in places... will be fun to play with.

Anyway, I've contributed zero to this, but thought I'd show a rendering from the software. It is pretty cool...

 

 

Edited September 5, 2021 by Andrew Davie

[+Andrew Davie]

I'm learning what all the code does, and how that affects the output.

Here I've turned off the error diffusion/stippling. Interesting!

 

 

 

[+rbairos]

Hey that's really cool. Looking forward to seeing what you find.

[+Andrew Davie]

Continuing with the experiments, I noticed that the algorithm prefers to select greys/whites instead of colours very similar to these. That is, would choose a grey instead of a pale green. So, I knocked out all the whites/greys from the palette to see how it would do. Also, I super-saturated the colours of the source before converting this one. I wanted to really enhance the variety of colours chosen. Finally, I chose this particular clip mostly because it has dinosaurs fighting, but also because it's incredibly messy/busy and should be very difficult to render well.
 

 

 

The results are pretty good, considering. The super-saturation is a bit of overkill, and perhaps the brightness should be down a bit. But the removal of the white/grey colours seem to be working well. I'll do another version with the original palette and without colour super-saturation just to show the difference. That will take a few hours to finish, though.

 

[+rbairos]

Really enjoying all the exploration you're doing.
What are you using to play with the saturation? There are endless ways to manipulate the color in the existing software.

Its interesting that it picks the greys instead of light greens.
This suggests that my color-distance function isn't optimal in that case.
Right now it just takes minimum euclidean distance, weighted by .6/.3/.1 for R/G/B
I have a color weighting method parameter to select the 8-bit cell color, but notice I didn't use it in

findClosestInPalette  / colorDist.
Maybe that will provide different selections that take hue into account better.

I know there are also other published methods for finding 'best' color in a set, that I never fully explored.

Happy exploring!

 

[+rbairos]

Also, I think I'm going to change the flicker/fill pattern from two alternating checkerboards to two alternating columns, to free up more cycles.
I had this originally and to be honest, the alternating checkerboards look like alternating columns in practice anyways.
I'll need the extra cycles to setup the playfield bits more readily for background, as the blank spots still need to be black, not background-color.

The only remaining advantage of checkerboard was that you still saw a pretty full frame, even if the youtube playback was 30hz instead of the intended 60hz. With columns, you definitely see only half the columns until you change that setting.

After that may re-explore the shiften chevron pattern Thomas Jentzsch proposes earlier.

 

Edited September 6, 2021 by rbairos

[Thomas Jentzsch]

6 minutes ago, rbairos said:

Also, I think I'm going to change the flicker/fill pattern from two alternating checkerboards to two alternating columns

You should be careful here, because this will increase flicker quite a lot.

[+Andrew Davie]

12 minutes ago, rbairos said:

Really enjoying all the exploration you're doing.
What are you using to play with the saturation? There are endless ways to manipulate the color in the existing software.

I ran the original through iMovie first. I don't know enough about the tool/existing software to use it competently.

 

12 minutes ago, rbairos said:

Its interesting that it picks the greys instead of light greens.
This suggests that my color-distance function isn't optimal in that case.
Right now it just takes minimum euclidean distance, weighted by .6/.3/.1 for R/G/B

I've just added a HSL conversion (which isn't functioning 100% yet - tomorow, though...) so that I can do a nearest-colour based on weightings on hue, saturation, lightness... rather than RGB.  Results tomorrow, hopefully.

 

 

[+Andrew Davie]

9 minutes ago, rbairos said:

Also, I think I'm going to change the flicker/fill pattern from two alternating checkerboards to two alternating columns, to free up more cycles.
 

Yeah, I think this is a terrible idea!!  It's going to be so totally noticeable it will hurt my eyes

 

[+rbairos]

1 hour ago, Andrew Davie said:

Yeah, I think this is a terrible idea!!  It's going to be so totally noticeable it will hurt my eyes

 

It still 30hz in both cases though. 
I'll make some test cases in emulation to compare.
The checkerboards ended up looking like columns anyways.
*Maybe* the chevrons though, but those end up creating swimming artifacts

If I do go with checkerboards, may have to just update the background color every second line.
Reason is I need to avoid the blank areas.

Fastest I can do this is by setting up playfield as    ----    ----    ----    ----    ----   ----
Then setting playfield color one line, while setting background to black.
Then swapping the colors for the second line.

The background color will extend through the left and right sides of the screen, but maybe this will look better,
and help hide the weird central rectangle viewing area I have now.

Okay I'll start with that. background every two lines on the original checkerboard.
 

Edited September 6, 2021 by rbairos

[+rbairos]

Quote

I've just added a HSL conversion (which isn't functioning 100% yet - tomorow, though...) so that I can do a nearest-colour based on weightings on hue, saturation, lightness... rather than RGB.  Results tomorrow, hopefully.

That's cool. Look through the original colorizer.cpp 
I have an RGBtoLum and RGBtoHue in there, but not really explored.

Edited September 6, 2021 by rbairos

[Thomas Jentzsch]

17 minutes ago, rbairos said:

It still 30hz in both cases though. 

Yes, but the checkerboard hides most of the flicker. There is a good reason why all flickering text kernels to this.

[+rbairos]

Just now, Thomas Jentzsch said:

Yes, but the checkerboard hides most of the flicker. There is a good reason why all flickering text kernels to this.

Okay fair enough.
 

[Thomas Jentzsch]

1 minute ago, rbairos said:

Okay fair enough.

That's why I was afraid that my idea from above would create too much flicker.

[Shawn]

flicker = bad

[+Andrew Davie]

Here's the original clip showing the tendency towards grey. I know the colour is subtle here, but I think the colour choice might be able to be improved slightly with a better matching algorithm. In any case, you can clearly see the differences when saturation is increased!

 

[+Andrew Davie]

Here's a version with colour saturation increased (50% of the first version from yesterday).

 

 

[+rbairos]

9 hours ago, Andrew Davie said:

Here's a version with colour saturation increased (50% of the first version from yesterday).

 

 

 

That's cool.
If you want to play with the saturation in realtime during the encoding, add a 'HSV Adjust' node in the chain.
The 'Saturation Multiplier' will do it.
 

[+rbairos]

Thing are moving along steadily on the kernel/emulator side.
I was able to find enough cycles by sacrificing the STA WSYNCs, for exact kernel entry times.
So maybe full background color change per line.
Fingers crossed.

[+Andrew Davie]

1 hour ago, rbairos said:

That's cool.
If you want to play with the saturation in realtime during the encoding, add a 'HSV Adjust' node in the chain.
The 'Saturation Multiplier' will do it.
 

That worked well, TY.

Why does the tool open two identical windows when I start it?  See image...

I close the top one, the one underneath keeps going.  Seems redundant.

 

 

[Thomas Jentzsch]

37 minutes ago, rbairos said:

Thing are moving along steadily on the kernel/emulator side.
I was able to find enough cycles by sacrificing the STA WSYNCs, for exact kernel entry times.
So maybe full background color change per line.
Fingers crossed.

The code I posted should provide enough free time.

[+rbairos]

22 minutes ago, Thomas Jentzsch said:

The code I posted should provide enough free time.

Yah absolutely, your earlier changes were key.
I was able to carve out a  colubk, and colupf  for the right kernel, and left kernel.
I need to change both colors, to keep the background color a checkerboard instead of solid.
Still need a few more changesto prove to myself this weeks simulations are actually TIA compliant, but hopeful.

My backup was to drop one half of the audio updates from 15k to 7.5K, but luckily didn't have to touch that.

 

[+rbairos]

29 minutes ago, Andrew Davie said:

That worked well, TY.

Why does the tool open two identical windows when I start it?  See image...

I close the top one, the one underneath keeps going.  Seems redundant.

 

 

I'll have to see how your specific file is now set up.
I had set it up to open only one performance window.
PM me for details.
 

[+Andrew Davie]

1 minute ago, rbairos said:

I'll have to see how your specific file is now set up.
I had set it up to open only one performance window.
PM me for details.
 

Will PM, but it's *always* done this for me.  I use the little white "STOP" square on the main window to launch it. Perhaps that's the issue.

 

[+Andrew Davie]

Here's the latest. I've played with saturation and tweaked a few other things. Perhaps too bright.

But I have to say I never thought I'd see this sort of quality on the '2600 for video.  It's astounding.

 

 

As an aside: YouTube detected a copyright claim/breach on the Thin Lizzy song "Boys are Back in Town" so I guess that says a bit about the audio quality - at least recognisable to an algorithm

 

Edited September 7, 2021 by Andrew Davie