Convergent Designs Flash XDR

[John Richard]

Was a great pleasure meetng you as well Dan.

Can't wait to see the sample footage - especially the 100mb format.

Do you know of any new info regarding NLE support for the various formats?

And thanks again for all the help you've provided in the H1 forum.

[Dan Keaton]

Dear John,

Thanks for the very kind words.

I do not have any updates concerning which Non-Linear Editors will support which bit rates at this time.

Mike will be addressing that issue. As we get closer to shipping, we will have some answers from the NLE vendors.

What I can say at this time, is that there is a dramatic increase in quality, based on our testing, between a 25 Mb file and a 50 Mb file.

More results of our image testing to come.....

[Mike Schell]

Originally Posted by John Richard

Was a great pleasure meetng you as well Dan.

Can't wait to see the sample footage - especially the 100mb format.

Do you know of any new info regarding NLE support for the various formats?

And thanks again for all the help you've provided in the H1 forum.

Hi John-
Thanks! Enjoyed meeting you also at NAB. It was quite a show for us and the Vidy and BE Pick Hit awards were icing on the cake!

I will be posting sample footage tomorrow. We have comparison images at uncompressed, 50 Mbps 422, DVCProHD and HDV rates. There is a very small difference between 50 Mbps and uncompressed. The 50 Mbps also clearly outperforms DVCProHD in our limited tests.

I'll try to get some 100 Mbps footage in the next few days. Our engineers are bringing up the final production board now and we hope to be writing MXF files to FAT32 partitions lalter this week.

I do think the 100 Mbps 422 Long-GOP will be the "sweet spot" for Flash XDR. I suspect that this rate will outperform even the 160 Mbps I-Frame only in most applications. Long-GOP is fundamentally a much more sophisticated compression algorithm compared to I-Frame only CODECs such as DVCProHD. By using both spatial (I-Frame) and temporal (P,B Frames) compression, Long-GOP clearly offers potentially higher overall quality. In my experience you have to increase the I-Frame bit-rate by 3 to 4X to match Long-GOP quality.

Long-GOP MPEG2 is more difficult to edit than I-Frame only, but this shortcoming is rapidly diminishing with faster multi-core processors.

It does look very likely that the low-cost 32GB Transcend CF card will support the 100 Mbps rate. I'll have more info in the coming weeks.

[Mike Schell]

Well, the long awaited images comparisons are finally starting to appear. Check out the latest downloads on the Convergent Design website home page. You will find comparisons of HDV, DVCProHD and 50, 100 Mbps as well as uncompressed images. In our view the 50 Mbps MPEG2 clearly outperforms DCVProHD, while the 100 Mbps is very close to uncompressed. (If you zoom in on the text, the differences become quite apparent).

In the coming weeks we will be working on green-screen comparisons as well as video with lots of motion.

You will also see some photos from NAB and our two awards.

[David Heath]

Originally Posted by Mike Schell

In our view the 50 Mbps MPEG2 clearly outperforms DCVProHD, while the 100 Mbps is very close to uncompressed.

On the file I downloaded, the comparisons appeared to be uncompressed, 25Mbs (HDV), 50Mbs long GOP, and 100Mbs I-frame only. (And no DVCProHD or 100Mbs long GOP.)

And on these images, the 50Mbs long GOP appeared superior to the 100Mbs I-frame only, though it would be interesting to see how motion would affect the comparison.

It raises an interesting thought in my mind. I've heard you say that all MPEG2 coders are not equal - an early coder and/or cheaper one would not do as good a job as a later, more sophisticated one, bitrate and other parameters being constant.

Is there an equivalent circumstance for DVCProHD (and DV for that matter)? Or does the codec itself uniquely define the compression quality?

[Mike Schell]

Originally Posted by David Heath

On the file I downloaded, the comparisons appeared to be uncompressed, 25Mbs (HDV), 50Mbs long GOP, and 100Mbs I-frame only. (And no DVCProHD or 100Mbs long GOP.)

And on these images, the 50Mbs long GOP appeared superior to the 100Mbs I-frame only, though it would be interesting to see how motion would affect the comparison.

It raises an interesting thought in my mind. I've heard you say that all MPEG2 coders are not equal - an early coder and/or cheaper one would not do as good a job as a later, more sophisticated one, bitrate and other parameters being constant.

Is there an equivalent circumstance for DVCProHD (and DV for that matter)? Or does the codec itself uniquely define the compression quality?

Hi David-
There are two comparison files: CODEC comparsion and BitRate comparison. The CODEC comparison includes HDV, DVCProHD, MPEG2 50 Mbps and uncompressed images. The bit-rate comparison includes HDV, MPEG2 50 Mbps & 100 Mbps and uncompressed.

Yes, there can be considerable differences in the quality of any video encoder be it DV, DVCProHD or MPEG2. All video CODEC specifications outline the requirements for the decoder (decompression), the encoder implementation (in either hardware or software) is left up to the manufacturer.

Over time, almost all CODEC encoders have improved with advances in silicon technology and better understanding of the algorithms. We consider the Final Cut Pro implementations to some of the very best software CODECs, hence they are used to compare aganist the Sony MPEG2 hardware CODEC used in Flash XDR.

[Dan Keaton]

Dear David,

There are multiple image download files on the www.convergent-design.com website.

Each of these has a "Read Me" file which explains the exact process that was used to create the various images.

So, it depends on which file you downloaded.

The "Fruit Image Comparison" file contains the following examples. This is included in the "Codec Comparison Images Close-Up" download.

1. A DVCPro HD example, which is 100 Mb intraframe encoding, using a Final Cut Pro encoder,

2. An HDV example, 25 Mb 4:2:0 Long GOP,

3. A 50 MB 4:2:2 Long Gop,

4. And an uncompressed 4:2:2 example.

It is important to download and view the full fruit image to realize that the "Cropped Fruit" file is only a very small portion of the full image, zoomed in to 300%. The full image, without any zoom in, is included in the "Codec Comparison Images" download.

A close examination of the zoomed in label on the Kiwi fruit shows the differences in the codecs.

This "Cropped Fruit" set does not include a 100 Mb 4:2:2 Long GOP example at this time.


Another set of examples shows the Convergent-Design brochure. I recommend that you zoom into the logo at the bottom of the page to compare each image. These are included in the "Bit Rate Comparison Images" download.

In this set of examples, the 100 Mb 4:2:2 Long GOP image is included so that you can compare it to uncompressed.

The Read Me files are very important as these attempt to explain the exact process that was used by Mike to create these files.

We are open to suggestions, comments and opinions.

[Jon Fairhurst]

Chris,

With all due respect for Convergent Designs Flash XDR, shouldn't this thread be moved out of the news section?

Every day I click on the News Forum to see the latest news, but it's usually just another XDR post.

[Chris Hurd]

Hmm -- is it time to move this thread (and the other XDR FAQ thread) out of news and into Tapeless?

[Mike Schell]

Originally Posted by Chris Hurd

Hmm -- is it time to move this thread (and the other XDR FAQ thread) out of news and into Tapeless?

Hi Chris-
This is the right general category, but Flash XDR is not direct to disk, per se. Maybe you could re-title this section as "Direct to Disk/Flash (Tapeless) Recording Solutions".

Just my 2 cents.

Best-
Mike Schell

[Thomas Smet]

Mike can we get a MXF sample from the device to check if it works in certain NLE's?

Also it might be a good idea to post a sample of something in motion or highly detailed. After all this is usually where 25mbits falls apart and this will be the area where people will notice the biggest advantage to the XDR. A still life shot will usually always look pretty good at 25mbits although clearly we can see that even for still shots higher bitrates do help.

[Mike Schell]

Originally Posted by Thomas Smet

Mike can we get a MXF sample from the device to check if it works in certain NLE's?

Also it might be a good idea to post a sample of something in motion or highly detailed. After all this is usually where 25mbits falls apart and this will be the area where people will notice the biggest advantage to the XDR. A still life shot will usually always look pretty good at 25mbits although clearly we can see that even for still shots higher bitrates do help.

Hi Thomas-
Good suggestions. As soon as we have done some basic verifications of our MXF file format, we will post example files for download and test. We have the MXF code finished, but we need to complete the FAT32 code first, which is very close.

I have some blue screen images to post next week and then we will test some full-motion video (Dan Keaton suggested panning around a vase of colorful flowers).

A couple of our users have done some very interesting tests with histograms and layer overlays in Photoshop. I will ask their permission to post some of these results on our website.

[John Richard]

Histograms that Mike noted.

Interesting in that the 50Mbps Sony/XDR 4:2:2 compares most closely to "Uncompressed" both visually when zoomed in for edges and detail and with their 2 respective histograms.

[Stefan Sargent]

Originally Posted by John Richard

Histograms that Mike noted.

Interesting in that the 50Mbps Sony/XDR 4:2:2 compares most closely to "Uncompressed" both visually when zoomed in for edges and detail and with their 2 respective histograms.

John,

Can you explain these histograms. I can see the differences but what are they really telling us? A "HISTOGRAMS FOR DUMMIES" ...

Thanks

Stefan

[John Richard]

Without meaning to sound like I fully understand this myself ....

In our area of interest in these single frame grabs from the 4 different formats shot:

A summary of a histogram in Photoshop is that it basically shows the distribution of all the pixels captured by the camera.

The horizontal scale goes from 0 on the left where total black is, to 255 on the right where total white is. The middle section is where the midtones of the image are displayed in the graph.

The vertical part of the graph displays how much of the image pixels are distributed in the blacks(shadow), midtones(grayscale), and highlight(whites).

So for example, if you had a shot that was totally underexposed, the histogram would graphicaly show most of the pixels distributed toward the right 0 end of the graph and very little in the middle and left 255 end of the graph.

If you had an overexposed or washed-out shot you would see most of the pixels distributed in the right end (255) of the graph.

But what is of most importance for what we are looking for in the study of the quality differences of the single frame grabs of these 4 different formats is how each format handled smoothly distributing the limited 8 bits of data in their respective historgram.

The scene shot is the same locked off shot and lighting. So what I was looking for was a comparison of the "smoothness" of the histogram - how much jagged combing or gaps there were - missing data due to the limits of 8 bit. These jagged gaps are where you get banding and aliasing - "jaggies" in edge detail that we all hate in digital video - it screams digital video. These gaps are what causes banding when we start applying effects or do some heavy compostiing. And it's why the capability to capture 10 bit formats instead of 8 bit is desireable.

In the 4 histograms provided for each of the respective formats, they all have gaps due to their 8 bit structure. But what I found interesting was how the 50Mbps Sony codec used in the XDR came closest to matching or emulating the 8 bit "Uncompressed" smoothness of pixel distribution. I think the historgrams are a statistical or numerical graphic display of what we are perceiving with review with our eyes and what we may expect when we get to review actual difficult footage with lots of movement in highly detailed subjects. I suspect that the smoother histogram of the 50mbps Sony codec recorded by the XDR is going to hold up better in these high motion/fine detail shots of flowing water or leafy trees in wind than that of the DVCProHD. I suspect that the 50mbps Sony/XDR will come closest to "Uncompressed" (hate that widely used misnomer).

I hope I made some sense and that the historgram comparisons are indeed valid.

For a much better and more in depth explanation of histograms, here is a link:

http://www.sphoto.com/techinfo/histo...histograms.htm

The other thing to keep in mind is the wonderful trickery of motion pictures vs. still photography - quickly flashing a succession of stills in front of our peepers hides a world of ills that may exist in the individual stills.