35Mbps 4:2:0 Vs 100Mbps 4:2:2 Comparison

[Mike Schell]

Courtesy of Jim Arthurs, we now have some definitive video comparisons of the Sony EX1 35Mbps 4:2:0 CODEC vs the Convergent Design Flash XDR (and nanoFlash) 100Mbps MPEG2 4:2:2 CODEC.

The test video was a fast horizontal pan, shot at 1/2000 shutter speed to eliminate any motion blur. This is a real torture test for a Long-GOP CODEC as each frame is different from the previous. The video was simultaneously captured on the SxS card (35Mbps) and on the Flash XDR portable recorder (100 Mbps), via the HD-SDI output. So, we have a true comparison of the two CODECs.

Zoom in on the images and you will notice an obvious improvement in the 100Mbps image, which is virtually 100% artifact free. IMO, the 100Mbps is visually lossless.

We need to rethink the need to always capture in I-Frame mode. Long-GOP MPEG2 is fundamentally a much more sophisticated compression algorithm, since it considers redundancies in both space (I-Frame) and time (P,B Frame). The rough rule of thumb is that 100 Mbps Long-GOP Quality = 250 Mbps I-Frame Only Quality.

Additionally, in other tests, we have seen that the 100Mbps Long-GOP MPEG2 Quality >> 100Mbps I-Frame only DVCProHD/JPEG2K Quality, even in high-motion conditions. In a simple resolution chart test, there is really no comparison of the Full-raster MPEG2 CODEC vs the DVCProHD CODEC in particular, since the DVCProHD CODEC subsamples the horizotal to 1280, while our MPEG2 maintians the full 1920 samples.

Here's the video: http://convergent-design.fileburst.com/35MbVRS100Mb_Video.mov

And here are the comparison images: http://convergent-design.fileburst.com/35MbVRS100Mb_Comparisons.zip

Mike Schell
Convergent Design

[Barry_Green]

No pan is a torture test for MPEG, because MPEG has motion prediction built in. If you really want to torture-test the codec you would want to rotate instead of pan, and throw in some color flashes and brightness flashes. Fireworks, or shooting people sitting around a campfire, fog & smoke, those are the things that cause long-GOP to choke. Constant motion is something that the motion prediction can compensate for, but brightness or color shifts or fog or flashes, those are all things that MPEG-2 doesn't have an answer for.

I would expect spectacular results from a 100mbps long-GOP 4:2:2, certainly. You should consider shooting the "Sirens of T.I." show in Las Vegas as a side-by-side or simultaneous torture test, I've found that show is a nearly "perfect storm" of conditions that cause long-GOP codecs to overload.

[Mike Schell]

No pan is a torture test for MPEG, because MPEG has motion prediction built in. If you really want to torture-test the codec you would want to rotate instead of pan, and throw in some color flashes and brightness flashes. Fireworks, or shooting people sitting around a campfire, fog & smoke, those are the things that cause long-GOP to choke. Constant motion is something that the motion prediction can compensate for, but brightness or color shifts or fog or flashes, those are all things that MPEG-2 doesn't have an answer for.

I would expect spectacular results from a 100mbps long-GOP 4:2:2, certainly. You should consider shooting the "Sirens of T.I." show in Las Vegas as a side-by-side or simultaneous torture test, I've found that show is a nearly "perfect storm" of conditions that cause long-GOP codecs to overload.

Hi Barry-
Good point, I agree. We can set up a very simple test by firing a flash from a still camera. But the "Sirens of T.I" would certainly be one of the ultimate tests.

I am betting this 7th generation Sony MPEG2 CODEC will hold up quite well, especially at 100 Mbps, but we will continue to test.

Mike Schell

[Mike Schell]

No pan is a torture test for MPEG, because MPEG has motion prediction built in. If you really want to torture-test the codec you would want to rotate instead of pan, and throw in some color flashes and brightness flashes. Fireworks, or shooting people sitting around a campfire, fog & smoke, those are the things that cause long-GOP to choke. Constant motion is something that the motion prediction can compensate for, but brightness or color shifts or fog or flashes, those are all things that MPEG-2 doesn't have an answer for.

I would expect spectacular results from a 100mbps long-GOP 4:2:2, certainly. You should consider shooting the "Sirens of T.I." show in Las Vegas as a side-by-side or simultaneous torture test, I've found that show is a nearly "perfect storm" of conditions that cause long-GOP codecs to overload.

Hi Barry-
Jim Arthurs, who is testing the Flash XDR with his Sony EX1 could not resist your challenge! So, he shot some fast moving rotational material, using a 1/2000 shutter speed to eliminate blurring. Once again he captured the video using the native 35Mbps 4:2:0 CODEC in the EX1 and simultaneously using the 100Mbps 4:2:2 CODEC in the Flash XDR (via HD-SDI). Note that both CODECs use the Long-GOP format.

I must admit you are correct on both points: 1) Rotating images are really CODEC busters, much more so than pans and 2) The 100 Mbps Long-GOP held up spectacularly.

Here:s the link to the video: http://convergent-design.fileburst.com/35MbVRS100Mb_Rotate.mov

Here's the link to the side by side comparisons: http://convergent-design.fileburst.com/35MbVRS100Mb_Rotate.zip

Mike Schell
Convergent Design

[Kholi]

Whoa. The Stills tell all! LoL Very nice, man. It makes the EX-1 codec look like garbage. I bet Jim Arthrus appreciates the Sony 422 Codec given how much VFX work he does.

Man can't wait to try this with the HPX170, although for different reasons.

[Stevet]

No pan is a torture test for MPEG, because MPEG has motion prediction built in.

Very true, but the stills from the comparision codecs are VERY revealing!

Thanks Mike and Jim Arthurs for the test, it looks great as expected!

[Stevet]

Whoa. The Stills tell all! LoL Very nice, man. It makes the EX-1 codec look like garbage. I bet Jim Arthrus appreciates the Sony 422 Codec given how much VFX work he does.

Man can't wait to try this with the HPX170, although for different reasons.


Kholi, I'd be real interested in those results! I know you have mixed feelings on the HPX170. That camera has amazing features. I'd really like to see a comparison on the DVCPRO HD codec verses the Convergent Design XDR (or nano) 100Mbps MPEG2 4:2:2 CODEC.

[Buck Forester]

Whoa, I see more of a difference in quality with the rotational comparisons than the horizontal pans! Very cool. Thanks for posting this stuff!

[booth]

It'd be very interesting to know if the Convergent Designs recorded is worth the money. Can't wait for more results to come in. 4:2:2 is something I miss from the HVX.

[Lor]

From one website I've seen, it cost over $4,000US for the Convergent Design Flash XDR. I'm not sure if a Mac Pro with a HD-SDI will cost that much.

[Stevet]

Wait for the NanoFlash which list at $3.5K, maybe under $3K street price?

[USLatin]

If you have to record at 100Mbps to get good results then that sucks... though there are other benefits on the HVX200 and HPX170, right? But they better be good for the additional $4,000 or more from the actual price and the nanoflash recorder needed or competitor...

Let me know if I missed or messed up something:

_ significantly sharper from the much higher pixel count of the sensor
_ added sensitivity from the size of the sensor
_ mechanic focus ring
_ ability to shoot 35Mbps 1080i (1080p for the EX3) and that's without the recorder

What am I missing?

[PerroneFord]

From one website I've seen, it cost over $4,000US for the Convergent Design Flash XDR. I'm not sure if a Mac Pro with a HD-SDI will cost that much.

You cant attach a mac to a set of rails on the camera...

[PerroneFord]

If you have to record at 100Mbps to get good results then that sucks...


Better tell Panasonic...



Let me know if I missed or messed up something:

_ significantly sharper from the much higher pixel count of the sensor
_ added sensitivity from the size of the sensor
_ mechanic focus ring
_ ability to shoot 35Mbps 1080i (1080p for the EX3) and that's without the recorder

What am I missing?

I really ha trouble following the logic of your post. But even then, I don't know what you mean by 1080p on the EX3. Both cameras record 1080p.

[editorforhire]

Barry,
Any rumors of a small-bodied Panasonic P2, 1/2" CCD camera to match the Sony EX1 or 3?

[PerroneFord]

Barry,
Any rumors of a small-bodied Panasonic P2, 1/2" CCD camera to match the Sony EX1 or 3?

You can't make a small body 1/2"CCD camera. Otherwise Sony would have done it. CMOS was the only way to make it work with current technology.

[Kholi]

If you have to record at 100Mbps to get good results then that sucks... though there are other benefits on the HVX200 and HPX170, right? But they better be good for the additional $4,000 or more from the actual price and the nanoflash recorder needed or competitor...

Let me know if I missed or messed up something:

_ significantly sharper from the much higher pixel count of the sensor
_ added sensitivity from the size of the sensor
_ mechanic focus ring
_ ability to shoot 35Mbps 1080i (1080p for the EX3) and that's without the recorder

What am I missing?

Not sure at all what you're talking about, but let me try...

1. EX1 and EX3 record to a Full Raster Sony Codec. They shoot a Full Frame 1080p unlike the HVX/HPX DVCproHD cameras.

2. The nanoFlash benefits the EX series Cameras by allowing you to bypass Sony's EX Codec. You get 4:2:2 and Better Compression. With the HPX170 it would allow you a full raster codec which means a very minor boost in resolved detail.

Is it worth it for either camera? That depends largely on what you plan to do with the material and if you can afford to shell out 4,000 (unit, cards, accessories) for it in addition to the amount of money you've already spent. I think it could be worth it for a number of people.

[Barry_Green]

Barry,
Any rumors of a small-bodied Panasonic P2, 1/2" CCD camera to match the Sony EX1 or 3?
Nobody's talking about any such product that I know of. Panasonic doesn't seem to think too highly of 1/2" chips, they had some SD 1/2" cameras before but now it's all 1/3" or 2/3". The HPX500 (after rebate) is already priced comparably to the EX3 so I don't expect them to do anything in that range.

[USLatin]

Better tell Panasonic...

But you get that into the cards. Big difference.I guess I didn't make myself clear.



Not sure at all what you're talking about, but let me try...

1. EX1 and EX3 record to a Full Raster Sony Codec. They shoot a Full Frame 1080p unlike the HVX/HPX DVCproHD cameras.

2. The nanoFlash benefits the EX series Cameras by allowing you to bypass Sony's EX Codec. You get 4:2:2 and Better Compression. With the HPX170 it would allow you a full raster codec which means a very minor boost in resolved detail.

Is it worth it for either camera? That depends largely on what you plan to do with the material and if you can afford to shell out 4,000 (unit, cards, accessories) for it in addition to the amount of money you've already spent. I think it could be worth it for a number of people.

Well that's just it Kholi, I am trying to read up about these cameras, which I know too little about. So you already did me a favor by pointing out some of the info and differences! Thanks!

Let me ask you a bit more. I saw and understand that the CMOS in the EX1 has a much higher sensor count, so I expect the SDI output from the EX1 to significantly sharper than the 170's image... either from the cards or the SDI.
But what exactly is full raster, and full frame? I want to make sure I understand this as clearly as possible.

[Fohdeesha]

Full raster conceptually means a 1 for 1 pixel mapping from the sensor all the way to the recording format. As in the sensors actually have 1920x1080 full pixels, that get recorded straight to a 1920x1080 format.

Quite unlike the hpx, who's sensors are something like 960x540, which get shifted, upscaled, etc to something like 1440xsomethingxsome odd pixel ratio DVCPRO HD, producing something like 500 actual tv lines of resolution.

I believe he meant the same thing by full frame, as all of the resolution is retained into the recording format, no cropping/downscaling

[USLatin]

aha... thanks for the clarification Fohdeesha!

so "native" is the same thing as "full raster" right? as in saying native 1080.

[JVR]

2

[USLatin]

what are you asking JVR?

[JVR]

3