Playing with the PS2 (Lets Get Crazy)

[Fudoh]

Doesn't the ossc require a video buffer in order to have "any" lag? Thought it was down to three scanlines or something....

yeah, that's BS. The OSSC is technically unable to produce any lag.

[strayan]

Another very simple form of deinterlacing similar to weave is to average the current and previous fields together, offset by one pixel. This has the same properties as weave (in terms of latency) but replaces the combing artifacts with one frame worth of ghosting. Which also may be considered less distracting than combing, depending on the content.

It is immensely less distracting! My ancient plasma TV has a ‘vertical filter’ setting which shifts the field by 1 pixel and it eliminates combing completely on 480i UltraHDMI N64 titles. The ghosting is barely noticable when things are in motion. Weave has been added to the PSDigital and will be with the PS2 version as well I believe.

[Fudoh]

It is immensely less distracting! My ancient plasma TV has a ‘vertical filter’ setting which shifts the field by 1 pixel and it eliminates combing completely on 480i UltraHDMI titles. The ghosting is barely noticable when things are in motion.

My Pioneers also shift the doubled fields against each other - forgot about those.

But blending? I don't believe your plasma is blending fields.

[Unseen]

By reducing the lag to zero (as suggested by both of you), you would always have to prioritize the current field.

Yes, that's what we proposed. We didn't say that it would look perfect, just that it might look better than pure bob or pure weave.

For example the XRGBs did already shift the doubled fields towards each other by 1 line, reducing the bob'ing effect rather a bit plus allowing for a complete restoration of 240p content being output in 480i

That sounds interesting, let's check if I understand what you mean. I'll call the lines from the first (top) field A1, A2 and A3 and the lines from the second (bottom) field B1, B2 and B3. Pure old-media-player-style weave would combine them like this:

A1
B1
A2
B2
A3
B3

GCVideo's bobbing would output two frames constructed like this (horizontal is consecutive output frames):

A1 - B1
A1 - B1
A2 - B2
A2 - B2
A3 - B3
A3 - B3

If I understand your description correctly, the old XRGBs would do this?

A1 - ? (black or B1 at top of frame, B0 if it exists)
A1 - B1
A2 - B1
A2 - B2
A3 - B2
A3 - B3
A4 - B3

[strayan]

It is immensely less distracting! My ancient plasma TV has a ‘vertical filter’ setting which shifts the field by 1 pixel and it eliminates combing completely on 480i UltraHDMI titles. The ghosting is barely noticable when things are in motion.

My Pioneers also shift the doubled fields against each other - forgot about those.

But blending? I don't believe your plasma is blending fields.

I took some quick and nasty footage here: https://youtu.be/1wPxVVoCYvw

It might not be blending but it definitely shifts the image by one pixel because if I enable scanlines and turn the filter on the scanlines disappear completely.

[Fudoh]

GCVideo's bobbing would output two frames constructed like this (horizontal is consecutive output frames): ...

Is this really the case? This looks already like shifted frames to me. This is what I would call "alternating between doubling up or down". IIrc this is what the XRGBs do and what restores a perfect stable image from titles like Dragon Blaze on PS2.

A1 - B1 *
A1 - B1 **

* marks the original "designated" position of Line A1, while ** marks the original "designated" position of B1.

While this

A2 - XX ***
A2 - B2 ****
XX - B2

*** marks the original "designated" position of Line A2, while **** marks the orginal "designated" postion of B2

is that the OSSC does and what leaves titles like Dragon Blaze with a up&down bob'ing effect.

Maybe it's the other way around, yes - - but in essence, yes, that's the difference. On an ideal deinterlacer I would like to have the manual option to shift the doubled frames against each other, like a -2, -1, 0, +1, +2 slider. Many years I took some raw interlaced captures of various PS2 titles, since I wanted to see what's the difference between titles that looked like crap on the XRGB and others that looked stunning. And in software I applied various pixel shifts after "hard-"doubling the fields. Was very interesting and useful. A completely underestimated deinterlacing option.

[Fudoh]

I took some quick and nasty footage here: https://youtu.be/1wPxVVoCYvw
It might not be blending but it definitely shifts the image by one pixel because if I enable scanlines and turn the filter on the scanlines disappear completely.

thanks! That's neat. Likely just a switch between doubling and weaving. When you say you enable scanlines (on your source?), does this make any sense on a 15khz output ?

[Unseen]

GCVideo's bobbing would output two frames constructed like this (horizontal is consecutive output frames): ...

Is this really the case? This looks already like shifted frames to me.

It should do aligned output unless I messed up something with the timing of the blanking area or padded active that introduces an artificial shift in one of the fields by moving the entire window.

I'm currently think that it's line-aligned because shifting the image by a line in every second output frame would require more memory for buffering. GCVideo currently has two line buffers, one of which captures the current input line while the other outputs the previous input line twice:

(input -> output)
A2 first part -> A1
A2 second part -> A1 again
A3 first part -> A2
A3 second part -> A2 again

Adding a one-line shift in B would need additional memory because B3 first half must be stored while B1 is still needed:

B2 first part -> (blank)
B2 second part -> B1
B3 first part -> B1 again
B3 second part -> B2
...

is that the OSSC does and what leaves titles like Dragon Blaze with a up&down bob'ing effect.

I remember having to fight with some bugs due to an incorrect active picture area during the development of GCVideo's line doubler that resulted in excessive flickering when applied to 480i content, maybe the OSSC hits the same problem or I accidentally "fixed" something in my code that is now incorrect, but looks better that way?

[Taiyaki]

I found the OSSC's bob-deinterlacing combined with the OSSC's scanlines filter turned all the way to 100% to look very similar to how 480i looks on a PVM (the PVM seen from very close).

I don't like any deinterlacing solution that attempts to give a progressive-like picture (Fudoh will hard disagree on this). For me, 480i looks best when it has the appearance of alternating lines that defines its appearance on the CRTs that interlacing was meant for. Combing artifacts? I don't mind them

I agree with that. Personally I think interlacing has its charm and it's part of our gaming history. If people want to re experience games the way they remember them then it's a must to leave 480i games as they were. The only time I argue it isn't ideal is when they changed games that originally ran in progressive into interlaced, at that point it kind of bugs me, like the bulk of the fighting game ports on the PS2, but on a good crt that handles it well it's not that bad at all.

[strayan]

I took some quick and nasty footage here: https://youtu.be/1wPxVVoCYvw
It might not be blending but it definitely shifts the image by one pixel because if I enable scanlines and turn the filter on the scanlines disappear completely.

thanks! That's neat. Likely just a switch between doubling and weaving.

I took some stills which give you a better idea what’s going on:

https://imgur.com/a/GEv4B7n

It’s an UltraHDMI outputting a progressive 480p frame from 480i game (so line doubled already) on a 480p plasma via DVI.

If someone can figure out what it’s doing and how to implement it some other way it really is the best way to handle 480i IMHO.

[Fudoh]

ok, but to be clear: your plasma doesn't even get a 15khz signal, so it's not applying any deinterlacing and it's not working a per-field basis. The vertical filter probably just blurs the frames. If you can find a source with a sharper output (like some small text), it's probably quite easy to tell what exactly is happening, when you enable the filter.

[Guspaz]

The thing is that right now, our only OSSC deinterlacing choice is bob. It seems like weave and interpolation could also be done without adding any lag (by the definition of delay in displaying the latest lines from the source) with very low implementation effort on the OSSC Pro. They just need a framebuffer to store the prior fields. I don't see what the harm in having bob/weave/interpolate modes for the deinterlacer on the OSSC Pro would be, some people may prefer them at all times, and they may be better suited to some content rather than others. My point was just that there is a lot more you can do with deinterlacing without any extra lag than just bob. Then, later, if people want to get fancy and try to implement adaptive solutions that use an algorithm to decide between temporal and spatial interpolation on a per-pixel or per-scanline basis, that's just a potential future bonus.

[strayan]

ok, but to be clear: your plasma doesn't even get a 15khz signal, so it's not applying any deinterlacing and it's not working a per-field basis. The vertical filter probably just blurs the frames. If you can find a source with a sharper output (like some small text), it's probably quite easy to tell what exactly is happening, when you enable the filter.

Have updated the link with sharper output.

What I am hoping is that the OSSC Pro is able weave deinterlace 480i content and then apply the same technique that my TV uses (vertical shift, blur, whatever it is) to the 480p output. No other method of displaying 480i that I have seen tops this because there is no latency cost.

Why gcvideo doesn’t handle 480i the same way the ultrahdmi does is beyond me.

[Fudoh]

sure thing. Bring'em all in Use case scenarios for weave are very limited though. You can apply it to 30fps material (basically like a PAL film mode, weaving with a 2:2 cadence), but the risk of losing the cadence lock is extremely high if you don't run a cadence detection in the background (which again would require a higher lag due to the extra buffer).

[kitty666cats]

The GBS8200 CFW also doesn't drop the frickin' image for 480i<->240p transitions, when I had an OSSC I never encountered a display that would stay sync'd up. Come to think of it, my LED Samsung smart TV handled it faster than any CRT monitors I tried it with @_@ but now I've got the glorious GBS

[Fudoh]

Have updated the link with sharper output.

Thank you! In the 2nd shot from the bottom you can see what the filter does, it's interpolating vertically by applying a 1px blur.
Exactly what I suggested a moment before your posting (last posting on first page). The difference is just that you're applying it on top of a weave filter while I suggested it by doing it on a bob'ed image. Does your UltraHDMI has other deinterlacing options than weave ?

[Fudoh]

The GBS8200 CFW also doesn't drop the frickin' image for 480i<->240p transitions

my Faroudjas don't either, but it comes at the cost of a framerate conversion. Good option to have though.

[Guspaz]

I actually rather think that a 60 fps weave would be preferable in a variety of cases. It would produce better results than bob in any low-motion game, like 2D RPGs, because it perfectly recreates a 480p image when there is no motion, and is not that distracting in low-motion. I suspect weave or interpolation would also be a preferred solution to bob in all cases for people who are sensitive to flicker. For those people, any visual degradation in the image may be offset by the bob flicker making them feel uncomfortable.

Weave has no problems locking to pulldown content, because there is no lock. You're interleaving every field, regardless of if it's properly interlaced or from a pulldown. You only need to lock if you're trying to reconstruct the progressive frame from before the pulldown.

[strayan]

Have updated the link with sharper output.

Thank you! In the 2nd shot from the bottom you can see what the filter does, it's interpolating vertically by applying a 1px blur.
Exactly what I suggested a moment before your posting (last posting on first page). The difference is just that you're applying it on top of a weave filter while I suggested it by doing it on a bob'ed image. Does your UltraHDMI has other deinterlacing options than weave ?

Cheers Fudoh, appreciate the trained eye. I could not quite figure out what is was doing. Applying the 1pixel vertical interpolation to the weaved image definitely looks good and completely eliminates combing. The UltraHDMI does not have any other deinterlacing options but I’m really happy with the default (weave) as long I can apply my TV’s 1 pixel blur. Really hoping this kind if weave+blur interpolation comes to the ps2Digital and ossc pro.

I’m now half expecting Extrems to blow in and tell me that SWISS can already do all this!

[strayan]

I actually rather think that a 60 fps weave or interpolation would be preferable in a variety of cases. It would produce better results than bob in any low-motion game, like 2D RPGs, because it perfectly recreates a 480p image when there is no motion, and is not that distracting in low-motion. I suspect weave or interpolation would also be a preferred solution to bob in all cases for people who are sensitive to flicker. For those people, any visual degradation in the image may be offset by the bob flicker making them feel uncomfortable.

Ding ding ding! This is me. Cannot cope with any flicker at all. Bob makes my eyes bleed.

[fernan1234]

There is a solder less solution in the works but you still.need to do some work to the board to clean up power and video.
I have built a few now for gamers to use pc crt on old consoles, lag tested and its ahead of OSSC over half a frame.
If anyone in Australia(or worldwide if you cant find someone closee) wants one done PM me.

Solderless solution sounds great. Other work on the board is fine as long as it only requires ability to follow instructions and everyday household tools.

But if that doesn't come to fruition I may hit you up even though I'm in the US, don't know of anyone over here offering these.

[Fudoh]

@strayan

would you - by any chance - be able to create a slightly longer video clip with the filter set to active (and without the actual TV menu on screen). Maybe your camera is even able to to a 60fps video in 720p? And maybe throw in a few more titles. It's a great show case for a totally lost feature. I wonder what the manfacturer's original intent was?

[Fudoh]

Weave has no problems locking to pulldown content, because there is no lock. You're interleaving every field, regardless of if it's properly interlaced or from a pulldown. You only need to lock if you're trying to reconstruct the progressive frame from before the pulldown.

but aren't you doing exactly this when dealing with 30fps material?

t = top field, b = bottom field

(T1) (B1) (T2) (B2) (T3) (B3) (T4) (B4)

T1/B1 weaved form a full frame without any motion between them.
B1/T2 weaved gives you combing errors.
T2/B2 weaved form a full frame again (and so on)

I mean, sure you can run a weave filter blindly across this, but it won't look any good. Weaving T1/B1, repeating the frame and then moving on to T2/B2 on the other hand gives you a combing-free 30fps output from an interlaced 30fps input.

Getting a stable film mode for PAL took the manufacturers YEARS to get right, since the detection is so much complicated than with 3:2 cadences. I don't a see a difference between 30fps interlaced material (basically NTSC 60Hz with 2:2 cadence) and pal film material (24fps sped up to 25fps and then displayed in PAL 50 Hz interlaced with 2:2 cadence).

I have processors with the option to set the deinterlacing to 2:2 weaving for NTSC (it's rare, but it exists). You can play the Playstation 2 conversions of Metal Slug 3/4/5 with this. Works perfectly fine.

[strayan]

@strayan

would you - by any chance - be able to create a slightly longer video clip with the filter set to active (and without the actual TV menu on screen). Maybe your camera is even able to to a 60fps video in 720p? And maybe throw in a few more titles. It's a great show case for a totally lost feature. I wonder what the manfacturer's original intent was?

The manual only makes two references to the purpose of the vertical filter both suggest it’s designed to reduce flicker (it doesn’t reduce bob flicker at all though) https://www.hitachi.com.au/documents/pr ... manual.pdf

It does appear to be a completely lost feature though. TV cost me AUD$25 and it has been the best TV I’ve owned.

I will do my best to get better quality footage for you in the next few days. The only video capture device I have is an iPad Pro and know close to nothing about shooting video. The last video I uploaded was in 1080p, why do you want 720p?

[Fudoh]

The last video I uploaded was in 1080p, why do you want 720p?

because you uploaded a 30fps video. Some cameras can do 720p in 60fps, but 1080p only in 30fps. Using an iPad Pro you can certainly shoot 60fps at any resolution, you just have to make sure to have it adjusted accordingly.

The manual only makes two references to the purpose of the vertical filter

I could imagine that it was meant for interlaced video and just "happens" to work on progressive feeds as well.

[strayan]

The last video I uploaded was in 1080p, why do you want 720p?

because you uploaded a 30fps video. Some cameras can do 720p in 60fps, but 1080p only in 30fps. Using an iPad Pro you can certainly shoot 60fps at any resolution, you just have to make sure to have it adjusted accordingly.

I’ll try and figure out how to do this.

Edit: on wow, I can even do 4k at 60fps! Had no idea.

Edit: footage requested https://www.youtube.com/watch?v=2Kn9_Qazn8w

[orange808]

@Fudoh

Had a look over the source this afternoon. The Gonbes exposes enough settings to make a difference. I doubt it will be identical to what you saw on the stock firmware.

[Guspaz]

Weave has no problems locking to pulldown content, because there is no lock. You're interleaving every field, regardless of if it's properly interlaced or from a pulldown. You only need to lock if you're trying to reconstruct the progressive frame from before the pulldown.

but aren't you doing exactly this when dealing with 30fps material?

t = top field, b = bottom field

(T1) (B1) (T2) (B2) (T3) (B3) (T4) (B4)

T1/B1 weaved form a full frame without any motion between them.
B1/T2 weaved gives you combing errors.
T2/B2 weaved form a full frame again (and so on)

I mean, sure you can run a weave filter blindly across this, but it won't look any good. Weaving T1/B1, repeating the frame and then moving on to T2/B2 on the other hand gives you a combing-free 30fps output from an interlaced 30fps input.

Getting a stable film mode for PAL took the manufacturers YEARS to get right, since the detection is so much complicated than with 3:2 cadences. I don't a see a difference between 30fps interlaced material (basically NTSC 60Hz with 2:2 cadence) and pal film material (24fps sped up to 25fps and then displayed in PAL 50 Hz interlaced with 2:2 cadence).

I have processors with the option to set the deinterlacing to 2:2 weaving for NTSC (it's rare, but it exists). You can play the Playstation 2 conversions of Metal Slug 3/4/5 with this. Works perfectly fine.

Nothing uses 30 FPS in practice, though. Film is ~24, PAL is ~25, television is ~60. Pulldown was done using unevenly repeating frame when splitting them into fields. Weave doesn't care about any of that, every frame output from a weave deinterlacer is going to have the current field and previous field interleaved into a progressive frame, kind of like how it would look on a CRT if you had even longer phosphor persistence. So there's nothing to lock on to, every field you receive is simply output as-is, using the scanlines from the previous field to fill the gaps. 480i60 in, 480p60 out. Or if you're dealing with PAL, 576i50 in, 576p50 out. Any mucking about that was done to content to get 24 progressive frames into 50 interlaced fields will simply get weaved together like it would have been displayed on the CRT.

The result of this is combing artifacts, of course, but pulldown was always an ugly hack to begin with. I'm also not sure how it's all that relevant, pulldown doesn't exist in the gaming world. It's only for video, which a retro scaling device isn't focused on. Yes, if you've got a video scaling solution, then there's a good incentive to support reconstructing progressive frames out of the pulleddown fields, but for retro game scaling, it's not important.

Here in North America, the most common form of pulldown was 3:2, which was used to fit 24 FPS film content (which is also commonly used for pre-recorded television programs) into 60 FPS interlaced TV broadcasts. This looks like this:



If you can lock-on to the pattern, you can reconstruct the input frames and convert a 60i signal into the 24p source losslessly. But my point is that, weaving produces a progressive output, and regardless of how good or bad it looks, it doesn't flicker.

[Fudoh]

But my point is that, weaving produces a progressive output, and regardless of how good or bad it looks, it doesn't flicker.

I don't disagree. My point simply was that by adding just a little bit of extra effort you can turn a mediocre experience on 30fps titles into a great one.

Nothing uses 30 FPS in practice

cone on. I'm not talking about the signal refresh rate, but the actual content. Of course there are 30fps video games. The Metal Slugs are great examples for 2D 30fps games and probably 50% of all current video games out there today run at 30fps.

Pulldown was done using unevenly repeating frame when splitting them into fields.

for NTSC, yes. For PAL it's an even 2:2 pulldown (25 to 50). That's what makes the detection in PAL so much more harder than for NTSC. And this is always what makes it so similar to interlacing on 30fps titles.

Weave doesn't care about any of that, every frame output from a weave deinterlacer is going to have the current field and previous field interleaved into a progressive frame

If you run a weave deinterlacing algorithm on a (with 2:2 cadence) on a 25fps/50i clip you get a 50% chance of getting it right. Give it a try. It all depends whether your first two fields form a proper frame or not. If they don't, your weaved result will be unusable. Unfortunately with analogue video sources of lesser quality your cadence won't be perfect throughout the whole run. This is why you not only need cadence detection, but you also need quick recovery from cadence breaks. Cadence detection was and still is a big deal after all.

480i60 in, 480p60 out.

I know what you mean, but I was trying to explain why a weaving deinterlacer can go terrible wrong for a 30fps game and it why it can turn out great instead with a little respect to cadence. Without a cadence detection you won't be able to influence what you get. Of course you can give a shit about it, but missing the cadence on 30fps/60i game is really not a good idea, when you have the chance to do it right. If you disregard cadence completely and instead of doubling frames output every weaved pair of fields (using every field twice), then you get 50% properly recontructed frames and 50% with massive combing. Since the ones with combing don't carry any information above the proper ones, you can just as easily drop them and duplicate the good ones instead.

but pulldown was always an ugly hack to begin with.

It's elegant.

I'm also not sure how it's all that relevant, pulldown doesn't exist in the gaming world.

if you think it doesn't, then try to explain to me how a 2:2 pulldown in video differs from what you see in interlaced 30fps games. Please!

the most common form of pulldown was 3:2, which was used to fit 24 FPS film content

I know, I can do it by hand and by heart - forward and inverse.


(PS: I'm not trying to battle you. I know what you're trying to say, but why negate any possibility of getting better results? Weaving on 30fps content is such a good example where paying a little attention to details can enhance the user experience greatly. You're looking for a one button solution for every source signal. I think the OSSC has long gone past that goal, so why not try to allow as many different options and settings as it requires to match any kind of signal to its best capabilities?)

[The_Guffman]

Wow, I had completely forgotten about this thread, when suddenly my phone starts blowing up with response notifications! I didn't think this would take off like this a month after I posted, but I'm glad I could inadvertently cause such an interesting discussion!

I'm definitely pretty interested with deinterlacing now myself. Since my original post when I didn't even know they existed, I've already picked up a few. I've also come to appreciate 480i a bit more, ESPECIALLY on a CRT as God intended, but Fudoh's right that there's an entire artistry to this that's almost lost to time. That plasma TV was pretty sweet to look at! If the OSSC Pro can do stuff like this, or will have room for it to be added in the future, I'm all for that. This is such a fascinating topic!