Codec Torture Test Anyone?

[Barlow Elton]

I have some very codec-challenging-but-real world footage shot with the XL-H1, recorded uncompressed HD-SDI to computer (via Kona/FCP) and also to HDV simultaneously. It's kids outside at dusk waving sparklers in front of the camera. My intent was to shoot something that totally overwhelms the HDV codec but is also a realistic scene to shoot.

Footage was shot at -3dB with no NR processing applied, in both 1080i and 24F modes.

I think it's the kind of footage that could really show the advantages of higher bit-rate HD recording.

I would love to see the uncompressed footage put through the Flash XDR and then see samples of how it compares to the uncompressed and HDV.

Let me know if anyone from CD is interested.

Barlow

[Barlow Elton]

I might be able to post some clips later (HUGE files!) but here are a few grabs from both the HDV recording and the live SDI.

There is also a 1600x900 interlace to progressive converted frame that I made using Compressor. I was actually able to get great looking overcrank slowmo from the SDI 1080i due to the lack of compression issues. It took awhile to render, but you would never know it came from 1080i footage.

Edit: The "overcrank" clip is up as a h264 QT. I encoded it with a very high bit rate to try and preserve the detail. SDI acquistion definitely helped get a superior conversion to progressive.

[Barlow Elton]

I think the other link has issues, this should work: http://homepage.mac.com/mrbarlowelton

[Ronan Fournier]

Thank you Barlow, that's exactly the kind of comparison we need.
Thanks to what Mike Schell said, I can imagine that if the test have been made with the Flash XDR @ 100Mb/s, the quality would have been very close of your HQ footage and far from the HDV, is that right?
The sharpness is much preserved in HQ.
That sounds promissing for the outside XDR footage test that Mike has anounced two weeks ago.

[Barlow Elton]

Thanks Ronan.

Obviously the video was shot to be trouble for a highly compressed codec. I wasn't trying to induce noise by gaining up, but by being slightly underexposed in very low light conditions, (the sky background is very difficult for HDV too) with only the sparklers lighting the kids faces. (there is a tiny Sony LED light hitting them from behind for a little separation)

The clips were shot with the 6x XL HD wide angle lens, wide open at F 1.6; the lens is definitely sharper with less CA when stopped down a bit, but I needed maximum light gathering.

My guess is that (in theory) the 100 mbs XDR mode should look nearly identical to the ProRes HQ clips.

Sorry for the pc folks, but the HDV files are Mac/FCP-centric. I could possibly post m2t's later.

[Barlow Elton]

Not sure if many pc users are aware of this, but ProRes can be viewed in QT on Windows now. Here's the link to the decoder--http://www.apple.com/support/downloads/appleproresquicktimedecoder10forwindows.html

[Barlow Elton]

Had to take down the clips. Limited bandwidth with the iDisk

[Mike Schell]

Originally Posted by Barlow Elton

Had to take down the clips. Limited bandwidth with the iDisk

Hi Barlow-
We should have the 100 Mbps QT version in the next day or so. We were battling 1080psf24 issues today, but should have that sorted very soon.

[Barlow Elton]

Originally Posted by Mike Schell

Hi Barlow-
We should have the 100 Mbps QT version in the next day or so. We were battling 1080psf24 issues today, but should have that sorted very soon.

Great! This should be interesting. :)

[Mike Schell]

Hi Barlow-
Our engineers (Brent, John and Tommy) got the 1080psf working today in QuickTime! So, we'll do the encode test with your Sparkler footage tomorrow and give you a call to work out the best place to post the files.

[Barlow Elton]

Wow. WOW WOW WOW!!!

Thanks Mike for sending me that converted clip.

I need to get some A/B pics up, but basically the 100mb mode REALLY DOES hold up visually to the uncompressed sample.

I blew up identical frames from both clips to twice their size and really put them through the microscope. I looked for trouble spots where the sparklers were nearly filling up the frame, and man, it's just as good as the HQ clip. You can still push and pull the images all over the place and not bring out troublesome artifacts.

I am a believer. :)

[Mike Schell]

Originally Posted by Barlow Elton

Wow. WOW WOW WOW!!!

Thanks Mike for sending me that converted clip.

I need to get some A/B pics up, but basically the 100mb mode REALLY DOES hold up visually to the uncompressed sample.

I blew up identical frames from both clips to twice their size and really put them through the microscope. I looked for trouble spots where the sparklers were nearly filling up the frame, and man, it's just as good as the HQ clip. You can still push and pull the images all over the place and not bring out troublesome artifacts.

I am a believer. :)

Hi Barlow-
Excellent news! It seems that Long-GOP MPEG2 holds up extremely well to CODEC torture tests, especially at the higher bit-rates.

The rule of thumb is that Long-GOP MPEG2 is roughly 2-3 times more efficient than an I-Frame only CODEC (such as ProRes, JPEG2K, or DNxHD). So 100 Mbps MPEG2 is approximately equivalent to 250 Mbps I-Frame in overall quality (ProRes HQ = 220 Mbps).

MPEG2 is actually is much more sophisticated compression algorithm since it not only considers redundancies within a single frame (I-Frame), but also from frame to frame (P,B Frames). MPEG2 got a bad rap with the bit-starved HDV format. As you crank up the bit-rate (and color resolution), MPEG2 becomes visually identical to uncompressed, IMO.

On final note regarding the playback performance of the full-raster 4:2:2 MPEG2 used in XDR/nano. Compared to HDV, I believe that our CODEC will deliver superior MAC/PC playback performance even at the higher bit-rates. HDV is 1440x1080, 4:2:0 MPEG2. So, when you decode (decompress) the video you have to expand every horizontal line from 1440 -> 1920 and then expand the color space from 4:2:0 -> 4:2:2. Our full raster 4:2:2 CODEC automatically decompresses to 1920x1080 4:2:2 frame, eliminating the need for this additional processing. We playback a 100 Mbps stream on our 3-year old MAC without any problems.

[Dan Keaton]

Dear Barlow,

Thank you for your post.

We have been most impressed with the 100 Mb Long GOP option.

Thank you for your confirmation of our internal tests.

"I love it when a plan comes together!", A-Team.

[Barlow Elton]

Originally Posted by Mike Schell

It seems that Long-GOP MPEG2 holds up extremely well to CODEC torture tests, especially at the higher bit-rates.

I had a feeling it would; it's really nice to see the proof now. :)

Originally Posted by Mike Schell

The rule of thumb is that Long-GOP MPEG2 is roughly 2-3 times more efficient than an I-Frame only CODEC (such as ProRes, JPEG2K, or DNxHD). So 100 Mbps MPEG2 is approximately equivalent to 250 Mbps I-Frame in overall quality (ProRes HQ = 220 Mbps).

Again, nice to see that (for once!) the theory is really true. The XDR's encoding in 100 mb is extremely good. I can say with a lot of confidence that 100mbit mode will be hugely popular with XDR/NanoFlash users, especially now with the lower cost and higher capacity of the latest CF cards.

Humongous difference compared to HDV. Heck, it's noticeably better than DVCPRO HD in my own testing.

Originally Posted by Mike Schell

MPEG2 got a bad rap with the bit-starved HDV format. As you crank up the bit-rate (and color resolution), MPEG2 becomes visually identical to uncompressed, IMO.

HDV is like Instant Coffee; Just enough for an HD kick, but definitely not the most satisfying flavor. I think with the XDR we will finally get a taste of some of the good Kona stuff. :)

Originally Posted by Mike Schell

On final note regarding the playback performance of the full-raster 4:2:2 MPEG2 used in XDR/nano. Compared to HDV, I believe that our CODEC will deliver superior MAC/PC playback performance even at the higher bit-rates. HDV is 1440x1080, 4:2:0 MPEG2. So, when you decode (decompress) the video you have to expand every horizontal line from 1440 -> 1920 and then expand the color space from 4:2:0 -> 4:2:2. Our full raster 4:2:2 CODEC automatically decompresses to 1920x1080 4:2:2 frame, eliminating the need for this additional processing. We playback a 100 Mbps stream on our 3-year old MAC without any problems.

Makes perfect sense. My G5 didn't have any trouble playing the 100mbs clip at all.