Best And worst Quality FMV (Streaming video) on the Sega CD?

[kool kitty89]

For time gal, (75*0x800-32552)/(20*24*32) would mean you could get a max of 7.8 fps just for the uncompressed tile data and 32KHz audio data. But if you start adding in palettes and palette maps, this optimal frame rate gets smaller, so ~6.3 seems appropriate.

Yes, but it's all just 16 colors per frame (so no color map and only 24 bytes per frame for CRAM updates) and the audio doesn't sound anywhere near 32 kHz (it doesn't even sound very good for 16 kHz, but that would seem likely -few if any games used 11 kHz iirc).
Very few examples of Japanese developed games used more than 1 palette for FMV. (Game Arts was one of the exceptions and, interestingly, did so with the very first example of FMV on the system -Tenka Fubu- which seems to be fairly similar to Sewer Shark and Night Trap)

It seems to be the exact same framerate and audio quality as Tenbu MCD Special, but with the frame cropped horizontally from 240 down to 192 pixels. (the intro portion also seems to use the same dithering scheme as Tenbu -different from Road Avenger, which has a lower color threshold and looks much more like typical floyd steinberg dithering)
Albeit, at that quality (240x160 16 colors 6.3 FPS), Tenbu would still have the bandwidth for 32 kHz PCM, but it definitely doesn't sound even close to that. (and that's assuming the data rate is actually maintained close to the 153600 bytes/s peak for mode 1 data)

So, given it's 16 colors per frame andassuming 16 kHz PCM, that would allow the same screen size and framerate as Road Avenger (7.5 FPS, or one update every 8 NTSC video frames), or a higher framerate at the 192x160 frame size. (doing 1/7th of 60 Hz would give ~8.6 FPS and should fit the bandwidth)

For Sonic CD FMV audio, I think the original FMV 32KHz 8 bit mono would suffice since 8 bit mono is the easiest to use for SCD FMV.

We were talking about trying to use paired channels for higher res (perhaps 13 or 14-bit) output via the ricoh chip here:
http://www.sega-16.com/forum/showthr...on-the-RF5C164

If that did work, you could do 8-bit ulaw, ADPCM decompressed to high res, or uncompressed high res PCM (though that would waste bandwidth). Of course, any such use of the ricoh chip would require double the buffer space in wave RAM (or more frequent updates of smaller buffers), so doing it Sonic CD style (32k double buffered and updated only once per second) would limit that to 16 kHz mono, but obviously you could do much more with smaller buffers updated more often. (albeit, at the expense of more overhead)


But yes, the existing 32 kHz PCM renditions are pretty good in any case.

[kool kitty89]

I had another look at the Sonic CD intro, and paid specific attention to what scenes were cut and what the duration of the cut sequences totaled to. It turns out to be approximately 18 seconds (that's seconds at the speed of the PC/etc versions) that are missing from the 90 second total, so the Sega CD version would be only ~72 seconds if those cuts were made directly. However, the video was slowed down to match the original duration, which would mean 80% of the original speed, or slowing a 30 FPS animation to 24 FPS and then cutting that down to 8 FPS by only using 1 of every 3 frames. (or from another perspective, they cut the source animation to 10 FPS by throwing out 2 of every 3 frames, slowed it to 8 FPS and cut out another 20% of the scenes used)

[BladeJunker]

3 years is a long time to bump something but it's related to FMV.

Was thinking about frame rate as a property of FMV quality and was wondering if halving the resolution height or width could fit in 2X the keyframes or save 50% CD space? Aesthetically double wide would look like C64, Mad Dog McCree 1 used this but not 2. Double tall pixels would look like something akin to Apple IIGS or PC88 graphics but 1/4 the res.

[BladeJunker]

Been a while so maybe people are like "What! Why would you do that, you crazy person, looks pretty bad already without you suggesting making it worse?" So maybe I need to give more context.

It's sort of a grading system based on the context of what you're showing and its priority when it comes to quality. Films would often save a little money dropping film grade for boring shots where nobody would notice much but on Sega CD you're saving on CD space. Wide angle shots tend to fair poorly with a res drop as was the case in Mad Dog McCree1 since distant people would be hard to make out when to shoot but in closeup shots of people and things where you were just watching a drop in res wasn't too bad.

Just an option really to not default to more and more CDs for such games which happened on PC post Sega CD where games could run 5 discs or more with swapping. Was just wondering if the option would work at all, been going over the ups and downs of FMV on Sega CD and thinking of good porting material in that field.

[bgvanbur]

You would save in the ballpark of 25 to 45% CD space if you use half the pixels. Since there is more to a video than just pixel data such as audio and file format overhead and less reuse going to a smaller display, you definitely won't get 50%.

Since the Sega CD is single speed and the Sega CD disc format says it supports 60 minutes of data (540 MB) it takes the Sega CD 60 minutes to read the entire disk at full speed. If you targeted full speed reads you would probably have buffering problems, so 70 minutes is probably a better target for the total video possible on a single Sega CD at full quality. I think changing the video size is more a functional of quality and a result of other graphics being displayed such as a GUI around the video since double buffered full screen video can easily use the entire VDP. Use the best quality and use multiple CDs if you got that much material Later gen systems could read discs much quicker and were more likely to not be limited by the disc read speed.

[BladeJunker]

You would save in the ballpark of 25 to 45% CD space if you use half the pixels. Since there is more to a video than just pixel data such as audio and file format overhead and less reuse going to a smaller display, you definitely won't get 50%.

Since the Sega CD is single speed and the Sega CD disc format says it supports 60 minutes of data (540 MB) it takes the Sega CD 60 minutes to read the entire disk at full speed. If you targeted full speed reads you would probably have buffering problems, so 70 minutes is probably a better target for the total video possible on a single Sega CD at full quality. I think changing the video size is more a functional of quality and a result of other graphics being displayed such as a GUI around the video since double buffered full screen video can easily use the entire VDP. Use the best quality and use multiple CDs if you got that much material Later gen systems could read discs much quicker and were more likely to not be limited by the disc read speed.

Thank you for the info. a low of 25% isn't much savings, less dramatic than I hoped lol. Right the VDP limit, why 63 color FMV doesn't work. No material, just theorizing. Mostly I'm just trying to improve the multimedia aspect of the Sega CD like transitions of 32 color FMV to stops which switch to higher color static shots and color palette concentration to improve the look of live action, come to the conclusion keying actors is the only way to gain control of there color saturation.

[Pyron]

Stellar Fire


Citizen-X is very well done for me...


Corpse Killer


Ground Zero Texas


Keio Flying Squadron - Mix Between streaming and PCE-CD stuff

[Pyron]

pce-cd like stuff:

Heroic legend of Arslan - kick your ass in @1:35


Bari-arm


Shadownrun


Shadow of the beast II


Dark Wizard

[Silanda]

Found something else that was interesting:




The top image is the original, obviously. It has familiar checker board dithering. The thing with checker board dithering, is that it only takes you up 1bit per RGB element. I applied a single pixel horizontal motion blur with a value of 1 (to the native res of the pic - 256x224) of checker board pattern dither, to get back the original values that the dither was trying to represent (with a few fringes artifacts, but they very minor). That's what pic 2 is.

The second pic had some left over horizontal lines (of the wider checker board dither). So, to figure out their color representation, I did another pass of motion blue, but this time 90 degrees instead of 0. This pic three. Pic 3 conforms exactly to a 15bit master palette. But you can still clearly see posterization. So the source image was probably 15bit color with a 256 color index (or possibly straight 15bit images).

Now all you have to do is hack some of these old FMV games and have the 32x apply a two pass motion blur (pixel accumulator) like I did and show it on the SuperVDP instead (Excluding Night Trap because that's already has a 32x version, right?)

Just a bump with regards to this. You can pretty much simulate this effect using a Framemeister. If H_Scaler is set to max (17?), and V_Scaler is set to 8, say goodbye to the dithering! In Digital Pictures games anyway; when I tested it, I didn't find it as effective on Tomcat Alley.

The output's blurry as hell but it looks decent for pure FMV games, and non-FMV games that were heavily dithered in expectation of non-RGB output may also benefit.