Quality expected from the EX3 and Flash XDR

[Steve Phillipps]

Looking at the cost of buying the quite exciting Sony EX3 and a Flash XDR box, you're getting upto around £11,000 or so, and close to the realms of serious gear like Panasonic HPX2100 etc. (well sort of 'cos I know you'd need a lens on top of that). Anyway just wondering what sort of quality you could expect from the EX3 and the XDR, close to (or better than?) DVCProHD from the likes of the HPX2100, Varicam, even Sony 750?
Any thoughts?
nd presumably this combo should be acceptable by broadcasters as full HD for programme acquisition?
Steve

[Thomas Smet]

In many cases the EX3 can already be of higher quality then DVCPROHD. It is only those few rare cases when DVCPROHD has the advantage. If you add the Flash XDR to the mix you get the option of I frame only recording at a bitrate much higher then DVCPROHD. If you choose long form recording of 100 mbits or I frame only at 160 mbits this pretty much blows away any recording format out there including HDCAM. About the only thing better is HDCAM SR but even then the long form 100 mbit should be pretty on par with HDCAM SR except for the lack of 4:4:4 RGB recording.

[Steve Phillipps]

I'd have trouble believing that an EX3 would provide better pictures than a Varicam, any real evidence that it might would be very interesting to see.
I was wondering (and I suppose doubting) whether EX1/3 +Flash XDR would even get close?
Steve

[Tim Polster]

I don't know the answer to your question, but I would be careful to make clear the difference between talking about codecs and talking about cameras.

The sentence - "In many cases the EX3 can already be of higher quality then DVCPROHD."

Could be misleading as the EX3 is a camera and DVCPRO HD is a codec.

People wil argue all day long that DVCPRO HD is better than MPEG-2 at 35 Mbps.

Ask yourself what is important to you? Resolution? Overall image impact?

I would guess that the XDR unit will help, but I would have to see how much.

[Mike Schell]

Originally Posted by Tim Polster

I don't know the answer to your question, but I would be careful to make clear the difference between talking about codecs and talking about cameras.

The sentence - "In many cases the EX3 can already be of higher quality then DVCPROHD."

Could be misleading as the EX3 is a camera and DVCPRO HD is a codec.

People wil argue all day long that DVCPRO HD is better than MPEG-2 at 35 Mbps.

Ask yourself what is important to you? Resolution? Overall image impact?

I would guess that the XDR unit will help, but I would have to see how much.

Hi Tim-
We have images on our website that compare the XDR MPEG2 50 Mbps 4:2:2 Long-GOP with DVCProHD. We used the Canon XL-H1 as our source in both cases. We captured directly into XDR and into the DVCProHD CODEC on Final Cut Pro.

In our view, the 50Mbps MPEG2 Long-GOP is clearly superior to DVCProHD. Part of the reason lies in the full-raster (1920x1080) vs the sub-sampling (1280x1080) employed in DVCProHD. The 100Mbps Long-GOP is very near uncompressed in quality.

It is always difficult to compare different cameras with different CODECs, but we think the EX1 + XDR in either 100Mbps Long-GOP or 160 Mbps I-Frame is going to produce incredible video, comparable to much more costly setups.

We will be testing this combination in the coming weeks.

[Tim Polster]

I agree, I am not trying to dispute.

I was mearly trying to point out that cameras and codecs are different.

Thomas' reply sounds as if the EX-1/3 are better than the DVCPRO codec.

The HPX series records DVCPRO HD 4:2:2 at 100 Mbps

As opposed to the EX1/3 which records Mpeg-2 4:2:0 at 35 Mbps

I wouldn't think a 4:2:0 color space would beat a 4:2:2 color space at any resonable bitrate.

On a side note:

Mike I went to the site and looked at the images.

I must say, I am looking forward to seeing some video before drawing any conclusions. Stills show some things, but video is a different animal.

Plus without seeing it on a production monitor or nice T.V., it is tough to make a call over the internet.

[David Heath]

Originally Posted by Tim Polster

I wouldn't think a 4:2:0 color space would beat a 4:2:2 color space at any resonable bitrate.

But that assumes equivalent luminance rasters, which here define "4". And that's not true if we compare the native EX1/3 codec with DVCProHD. The former is 1920x1080, DVCProHD is 1280x1080, so hence chrominance rasters of 960x540 versus 640x1080. And certainly for progressive recording I'd expect 1920x1080 4:2:0 to outperform 1280x1080 4:2:2, all else equal.

But I do agree that cameras shouldn't be confused with codecs. Given identical source material (as in Mikes tests) MPEG2 may outperform DVCProHD - and the 50Mbs variants upwards are 4:2:2 themselves - but different cameras mean different source materials.

[Tim Polster]

Wow, this stuff can get complex when all of the variables are taken into account.

I think I will just listen for a while as I need to learn more about this space.

[Chris Hurd]

Topic changed from "A different Flash XDR topic"

to "Quality expected from the EX3 and Flash XDR"

-- please avoid ambiguous thread titles on this site.

[Steve Phillipps]

Same here Tim!

[Dan Keaton]

Dear Steve and Tim,

We started our testing with a video image, created by a static image, using a locked down camera, in other words, no motion.

We did this for a number of reasons, but foremost was that this was a great way to show the very detailed differences between the quality of the codecs, using the exact same image as the source.

When viewing video at 24 frames per second, or 30 frames a second, small imperfections in the image are sometimes difficult to see.

We fully realize that posting full motion video is very desirable and we will be doing so in the near future.

Mike Schell and I have been discussing the best way to accomplish a meaningful comparison of full motion video.

We have, in house, a very nice Canon XL H1 and Canon XH G1. Both produce high quality uncompressed images through the HD-SDI port.

My suggestion to Mike, for the first round of full motion video testing, was to get a very nice floral arrangement, and then pan around while zoomed in on the intricate parts of the flowers.

Our challenges include the fact that we want to do this in a studio setting (Convergent-Design's office) since we are recording uncompressed HD-SDI to a Macintosh, using Final Cut Pro. We can go outside, if we get a longer HD-SDI cable. One option, once we get a longer cable, would be to record cars passing by. I believe that we could accentuate the movement by recording at high zoom ratios.

In any case, for these early tests, we want to show the differences between the various codecs such as HDV, DVCPro HD and uncompressed. Also, we feel that it is useful for us to show the differences between Long GOP and I-Frame only codecs.

We felt that MPEG2, Long GOP at 25 MB maximum bit rate would benefit from a higher bit rate encoding.

I did some HDV torture testing when I first purchased my XL H1. I intentionally recorded images that I thought it could not handle and found it could handle at lot more than I ever expected. But, I could find a few cases where it did have some problems.

Personally, while I believe HDV is almost a miracle in how well it actually works, it can benefit from a higher bit rate. So we used HDV (MPEG2 4:2:0 at 25 Mb) as are baseline for all comparisons.

Our results show that this is true, (but note that we also changed from 4:2:0 to 4:2:2 for our tests at higher bit rates.)

Switching from 4:2:0 to 4:2:2 is a very nice side benefit from using the HD-SDI port on most (if not all) cameras. Since we can record 4:2:2 using the Flash XDR so easily, we are doing so in our testing.

So, we welcome your suggestions as how we should proceed with our testing of full motion video.

At this point in our testing, we want to use one uncompressed HD-SDI image, as the basis for our codec/bit rate testing.

In the future, we will be comparing the outputs from various cameras, using their HD-SDI outputs, recording directly into our Flash XDR using various recording options.

[Tim Polster]

Hello Dan,

I think motion and Color fidelity are at the top of my list to see in a comparison video.

I have read about some problems with 720p60 and artifacts in the JVC forum, so maybe some testing at that framerate would show some weakness in the HDV spec.

Overall as a potential consumer, I need to justify the increased image quality against almost the cost of another camera ($5,000) in purchasing the Flash XDR.

The amount of the noticeable image quality increase vs the price of memory for the camera will make my decision.

But I have to be honest, as camera memory prices fall, the case for the second camera becomes stronger :(

Mainly because DVD is still the main delivery method and a lot of the HD goodness may be lost in translation...

[Mike Schell]

Originally Posted by Tim Polster

I agree, I am not trying to dispute.

I was mearly trying to point out that cameras and codecs are different.

Thomas' reply sounds as if the EX-1/3 are better than the DVCPRO codec.

The HPX series records DVCPRO HD 4:2:2 at 100 Mbps

As opposed to the EX1/3 which records Mpeg-2 4:2:0 at 35 Mbps

I wouldn't think a 4:2:0 color space would beat a 4:2:2 color space at any resonable bitrate.

On a side note:

Mike I went to the site and looked at the images.

I must say, I am looking forward to seeing some video before drawing any conclusions. Stills show some things, but video is a different animal.

Plus without seeing it on a production monitor or nice T.V., it is tough to make a call over the internet.

Hi Tim and Everyone-
These are very interesting discussions! I think everyone has now concluded that it's extremely difficult to compare different cameras with different sensors with different compressions CODECs. DVCProHD I-Frame (1280x 1080) 4:2:2 may or may not be better than MPEG2 Long-GOP (1920x1080) 4:2:0. On paper, it's a difficult call and may depend on the actual content.

But, I think we can continue to compare these various CODECs by using a single camera source (Canon XL-H1 or Sony EX1, for example) and capturing into these various CODECs. We can dial in the 35 Mbps EX1 CODEC in Flash XDR for these tests.

We can also do some comparisons of the front-end of some of these cameras by capturing the uncompressed HD-SDI output. This should allow us to separate the camera front-end from the native CODEC.

I am excited to report that the new Transcend 16GB 300X card is now available for around $240. This should support the 160 Mbps I-Frame only mode (about 13 minutes storage per card or 52 minutes for 4 cards). The 160 Mbps I-Frame only mode should yield some incredible results for 1080p24, since this is a full-raster (1920x1080) 4:2:2 8-bit capture with an effective compression of 5:1. This should be visually lossless and of course, no motion artifacts from Long-GOP.

I would still contend that for most applications, the 100Mbps Long-GOP (1920x1080) 4:2:2 is the "sweet spot". This rate is supported by the lower cost 32GB cards ($150) and should be very close to visually lossless. The only downside of this format occurs during very high-motion or fast changing scenes.

But Flash XDR gives you the option. Use 100Mbps Long-GOP for most of your shoots, then switch to 160Mbps for green-screen and high-action.

We will be posting video shot with both of these formats in the near future.

Also, as side note, I got a recent e-mail reminding me that Flash XDR can also be used to accelerate the encoding of Blu-Ray material. You should be a able to feed HD-SDI (from your NLE) to XDR and encode at 18-20 Mbps 4:2:0 in real-time. Then read the compressed footage (from the CF cards) back into your Blu-Ray authoring program. This should save you a ton of time in rendering the content.

[David Heath]

Originally Posted by Dan Keaton

So, we welcome your suggestions as how we should proceed with our testing of full motion video.

Firstly, I believe there are "reference" uncompressed video samples used by bodies such as the EBU, specifically to test various codec aspects. Some specifically test motion performance, some fine detail, others colour detail etc. I don't know if they are available to third parties, if not, your approach of generating a reference via HD-SDI, and deriving all the other versions from that uncompressed recording seems a sound approach.

In another thread, the technique is discussed of using Photoshop in difference mode, and using the filmstrip option extends making true comparisons from stills to moving images.

An example would be a clip of video, say 50 frames long, and lets say it consists of 15 frames of stationary image, 20 frames of panning over fine detail, and a further 15 frames of stationary image. An HDV (say) version is derived from the uncompressed original. Both versions are brought in to Photoshop as filmstrips, and separated into separate Y,U, and V. A given frame from the HDV version should be subtracted from the equivalent uncompressed version, and mid grey added.

If all the results were plain grey frames, the compression should be considered perfect. In practice, imperfections and the results of subsampling will show up as deviations from uniform grey.

If now frames from the first, second and third section are compared, and the deviations are found to be far greater for the second compared to the other two, it may be taken as a sign of how much the codec is "breaking down" with motion.

[Dan Keaton]

Dear David,

Thank you for your excellent suggestions!

Since HDV-II, Canon's Flavor of 25 Mb MPEG2 Long GOP recording, uses 15 frame "Group of Pictures", which is about 1/2 a second, we should have each segment be at least 1 to 2 seconds long so that we have a complete "Group of Pictures" for each segment.

David, your suggestions seem to be very sound. Our goal is to have the best possible comparisons.

When we shoot the motion video footage, we will be posting the footage on our website so that others can perform their own tests, in addition to the tests that we will be performing. Of course, the real problem is posting uncompressed footage due to the size of the files.