Intensity HV20 Footage

David Newman · May 15th, 2007, 02:52 PM

Sorry, I got the path wrong, this works : http://www.miscdata.com/downloads/NEOHDV1920.png

David Newman · May 15th, 2007, 02:57 PM

Quote:

Originally Posted by Derek Green

David, sorry I can't get your link to work. How do you recommend converting 1440 to 1920 for highest quality? I was putting the footage on a 1920 square pixel comp in aftereffects and then exporting the image. I'd like to redo my screengrabs if there's a better way.

It looks like you had an interlaced composite be so the fields were blended during the rescale. Note : this doesn't explain why the MJPEG is missing the top and bottom scanlines -- how did that happen?

Derek Green · May 15th, 2007, 03:02 PM

Quote:

Originally Posted by Noah Yuan-Vogel

The login worked fine for me. Certainly interesting seeing this comparison. Thanks for going to all the trouble to provide us with those samples. I've only looked at the tiffs since I havent had time to download the videos. For some reason, MJPEG seems to have some interlaced artifacting, but none of the other samples seem to... What was your workflow for arriving at these. Why would there by interlaced artifacting on only the MJPEG? I imagine that wouldnt help the compression artifacting since it introduces some extra fine detail. perhaps the blackmaging MJPEG is interlaced-only? compressing in fields even on progressive video? was the original uncompressed footage inverse telecined to 24fps before all compression?

Sorry guys, I was going to post more details about the workflow but I was trying to get this up before I had to go out for awhile.

Details:

-Camera was 4ft away from objects
-No zoom
-TV mode 48 frame rate; 24p mode
-Manual white balance; dropped exposure -2 to get rid of any zebras
-Removed pulldown on the NEOHD and NEOHDV in real time capture. You will see interlacing on these two for the first 10 frames or so, as this is what has been talked about in other posts.
-I didn't attempt to remove pulldown on any of the other clips. They should all be interlaced but you should be able to remove pulldown as the camera was set to 24p.
-all blackmagic clips were captured with their capture utility
-all cineform clips were captured with hdlink
-all clips were brought into a square pixel 1920 comp in after effects. Like I mentioned earlier, after effects flips the MJPEG video for some reason so I had to flip it.
-exported as a photoshop file from after effects
-exported as tiffs from photoshop

Noah Yuan-Vogel · May 15th, 2007, 03:47 PM

So if only the cineform ones were inverse telecine'd, theres a good chance the MJPEG one was still telecined and the screen grab ended up happening on a hybrid frame? maybe thats why there seem to be interlace artifacts. David's rescale of the neo hdv frame certainly does look much better than the previous version of that image. Any idea why it's so different?

Derek Green · May 15th, 2007, 03:49 PM

Quote:

Originally Posted by David Newman

It looks like you had an interlaced composite be so the fields were blended during the rescale. Note : this doesn't explain why the MJPEG is missing the top and bottom scanlines -- how did that happen?

Sorry David, your observation is correct. I went back and checked my AE file and for some reason, the NEOHDV footage had separating fields on. I have replaced the PSfile and TIFF file of the NEOHDV footage and updated my original post.

I'd like to retract my observations that I posted in the initial post about NEOHDV. It does look a lot better then the MJPEG. I'm glad I was wrong about this, as I was really hoping in the first place that the NEOHDV would be a good option since it doesn't cost all that much.

David Newman · May 15th, 2007, 03:59 PM

Thanks Derek. That is what we where hoping people would find. While there is a small cost to NEO HDV, the benefits of improved quality and on the fly pulldown removal, makes it idea for a low cost, "uncompressed-lite", capture and post solution of these new cameras.

Mike Thomann · May 15th, 2007, 09:06 PM

I made some comparisons of Derek's footage and found what to me looks like very positive results as to whether or not 1920x1080 is downrezzed to 1440x1080 and back to 1920x1080 before HDMI output. I used Photoshop to zoom in 1600% on a target area of the HDV sample which we know has been through the down-conversion process due to the HDV codec. As I examined the pixels, both closely and at a distance, I saw vertical lines. I highlighted them to reveal that they were evenly spaced at every 4th horizontal pixel, which happens to correspond to pixel interpolation of every 4th pixel associated with a 25% horizontal pixel contraction and expansion in the conversion process.

Then I examined the same area on Derek's uncompressed sample and did not see any evidence of vertical patterns, which is a very good thing because that means down-conversion before HDMI output is not evident within these samples.

I uploaded the highlighted comparisons to http://thomann.net/hv20/interpolation/

EDIT: What is most notable about these comparisons is that each pixel within the uncompressed sample sharply appears to represent a single individual color on its own, especially when compared to HDV. That is some incredible footage. I am extremely impressed.

EDIT: I have also examined NEOHDV against NEOHD and Uncompressed. As expected, since NEOHDV is 1440x1080, the same lines are evident every 4th horizontal pixel; whereas they are not evident with NEOHD and uncompressed which are full 1920x1080. This is further evidence that HDMI out has not been down-converted to 1440x1080. I have added these comparisons to the link above as well.

Terence Krueger · May 16th, 2007, 12:11 AM

"This is further evidence that HDMI out has not been down-converted to 1440x1080. I have added these comparisons to the link above as well"

it could also just mean that the camcorder is better at resampling the footage than whatever editor derek is using since the hdv and neohdv are actually 1440x1080 on disk.

i think the footage looks great regardless.

terence

Mike Thomann · May 16th, 2007, 12:19 AM

When having to remove and re-create every 4th pixel, the only option is to take an average of the pixels around it. The only variation is in which pixels you take an average from. Photoshop provides a choice of nearest neighbor, linear, and bicubic. I have resized the uncompressed sample to 1440x1080 and back to 1920x1080 with Photoshop using those three resampling methods. I then did a layered comparison of those to the original uncompressed sample. Bicubic and linear were nearly identical to each other as expected in a strictly horizontal resize and both appear to impose quite a bit of blur on the image on the pixel level, and without comparing to their original source there are no signs of downrezzing. After looking at them, the uncompressed appears 3D like and razor sharp. The nearest neighbor resampling method on the other hand showed the same pronounced evidence of downrezzing as HDV and NEOHDV, but this method appeared just as sharp as the uncompressed but with less distinguishable detail. It may stand to reason that *IF* the HV20 was to downrez before HDMI output, it would do so with this method and hence we would see evidence of it. There is a chance that the HV20 uses a resizing method such as linear or bicubic which would leave little evidence of resizing. When looking at the amount of blur they generate on the pixel level I'm unable to see how Derek's uncompressed sample could possibly be the result of such a method. There's just too much blur taking place that spans all the pixels to the point that it appears as though you're looking at a compression codec. I'm not typically one to come to such conclusions but it's obvious to me that Derek's uncompressed sample has not been through such a process, and I'm happy with that. In the end, even if resizing was taking place, at these resolutions I need to view it a foot away with at least 300% zoom to begin to notice and in the real world I would need to be within 10 feet of a 55 foot screen. ...but like I said I see it as obvious that no resizing is even taking place.

Glenn Thomas · May 16th, 2007, 05:41 AM

David, I'm quite impressed by your Neo HD codec, but at $599 its more expensive than a copy of Vegas+DVD, and is also the same price as a MacMini. At that price I just think you could be excluding a large portion of low budget indie film makers who are just starting out. If they're working with the HV20 for instance, they're going to need a fast computer to start with which may cost as much as the camera. Then an Intensity card if they plan on capturing high quality (1920x1080P) footage using HDMI, which I've seen for $250. Finally the NEO HD codec which is more than double the price of the Intensity.

Don't take this the wrong way as I realise it's a great product and I'm sure you'll sell a lot to more professional film makers, but a lot of people coming from say a DV background just aren't going to be able to fathom paying so much for what is essentially a capture codec. Perhaps you could do a cheaper cature only version without all the conversion features?

Thomas Smet · May 16th, 2007, 07:58 AM

Glenn, High quality comes at a cost. That is why there is NeoHDV which is much cheaper. If you are really on a budget you could always use the free mjpeg codec that comes with the Intensity card.

In my opinion the fact that you can now get a camera that can shoot 24p for $1,000.00 compared to in the past needing to buy a DVX100 for over $3,000.00 should mean many indies should have about $2,000.00 left over to buy a new computer and software. Everything is already much cheaper then it used to be. With SD 24p many of us had to spend a lot of money to get higher then DV quality. Even if you wanted to capture live uncompressed the only option we had was a Y/C S-video output from our cameras. The only camera I had that could pump out close to uncompressed SD was a $10,000 DSR-300 with a $80 26 pin connector that allowed component output. Now for $1,000.00 + $250.00 we can capture pure uncompressed or lightly compressed HD.

There is also the fact that there really isn't all that much extra detail when you capture 1920x1080 from pretty much any HD camera. You will only notice a very tiny amount of sharpness between 1920 and 1440. If somebody really insists on having that tiny edge in detail then they should be ready to pay for that tiny boost.

Wayne Morellini · May 16th, 2007, 08:45 AM

Quote:

Originally Posted by Derek Green

So I had a chance to shoot some greenscreen tests tonight with the Intensity and will be posting the video and pics asap.

My initial observations of the different shooting options I tried:

1. Uncompressed

No brainer, it looks the best, but not over NEOHD by as much as you'd think. Unfortunately it's just not an economic reality to shoot uncompressed. You'd probably be much better off saving all the money you'd spend on storage and buying a mini Red Camera when they come out. I don't have a raid setup so I was only able to get about 8 seconds of video before the recording stopped.

2. NEOHD

Looks really, really good. Gives the uncompressed footage a real run for it's money. Seriously, you have to blow it up to 200% to see any real difference and even then it's just a fraction of a bit more smudgy then the uncompressed but to even see that little difference you have to squint at a spot and flip the images back and forth.

If it is obvious at 200%, then it might stick out in an theatre.

NeoHD, something new.

Has anybody tried the trial version of this lossless codec, averaging over 6:1 compression:
http://www.digitalanarchy.com/micro/micro_main.html

I understood, from BM, that Intensity was supposed to allow you to sue any codec you wanted?

Thanks for the images, I am starting to download them.

Wayne Morellini · May 16th, 2007, 08:51 AM

David, for your codecs and software, can I suggest an cleanup/film processing tool. Basically cleans artifacts, restores image, removes noise from compressed images like XDCAMHD HDV, MiniDV, AVCHD, h264 cameras (like the recent 720p $299 h264 pocket camera, and Sanyo in news). Both applied post, and automatically on capture. Back to HV20 Intensity.

David Newman · May 16th, 2007, 10:08 AM

>If it is obvious at 200%, then it might stick out in an theatre.

He didn't say it was obvious, the compression is light only subtle noise variation is visible in magnified views. We aim, just like uncompressed, to store everthing, noise and all. Also theatre presentations are easiest to achieve a quality result, we passed all those test years ago. Just consider RED's snafu, the 4K presentation looked great, yet when people got to see the stills they saw compression issues, which is an issue for effects work, now the camera is delayed. It is much harder to do compression for the compositor -- this is the area we have grown. Read more here : http://cineform.blogspot.com/2007/04...eform-444.html

Anyway, aren't we talking about the HV20 here? Are we really planning theatre presentations? :)

> Has anybody tried the trial version of this lossless codec, averaging over 6:1 compression

Lossless compression only averages 2:1 from any vendor, the noise component of film or video is too great for much more compression. Your idea (which we discussed before) of extracting the noise, and compressing only the "signal" is an idea that can't be reached to the industries satisfaction -- always some signal is lost with the noise reduction. This is why highend is so pro uncompressed, signal + noise, decide later what to do with it. Yet uncompressed is such a pain, and the industry has generally accepted HDCAM-SR level compression, this is the compression quality target for us: store everything such that you couldn't tell which was the source. More here : http://cineform.blogspot.com/2007/03...en-screen.html

Now back the the HV20 + Intensity thread.

Noah Yuan-Vogel · May 16th, 2007, 11:06 AM

from the sound of it, that microcosm codec might be too slow for realtime capture, and isnt much more better at compressing than bitjazz sheervideo. Looks like for live video it only gets around 3:1 compression and only supports RGBA.

May 15th, 2007, 02:52 PM	#16
David Newman CTO, CineForm Inc. Join Date: Jul 2003 Location: Cardiff-by-the-Sea, California Posts: 8,095	Sorry, I got the path wrong, this works : http://www.miscdata.com/downloads/NEOHDV1920.png __________________ David Newman -- web: www.gopro.com blog: cineform.blogspot.com -- twitter: twitter.com/David_Newman

May 15th, 2007, 03:59 PM	#21
David Newman CTO, CineForm Inc. Join Date: Jul 2003 Location: Cardiff-by-the-Sea, California Posts: 8,095	Thanks Derek. That is what we where hoping people would find. While there is a small cost to NEO HDV, the benefits of improved quality and on the fly pulldown removal, makes it idea for a low cost, "uncompressed-lite", capture and post solution of these new cameras. __________________ David Newman -- web: www.gopro.com blog: cineform.blogspot.com -- twitter: twitter.com/David_Newman

May 15th, 2007, 09:06 PM	#22
Mike Thomann Regular Crew Join Date: May 2007 Location: Amherst, NY Posts: 100	I made some comparisons of Derek's footage and found what to me looks like very positive results as to whether or not 1920x1080 is downrezzed to 1440x1080 and back to 1920x1080 before HDMI output. I used Photoshop to zoom in 1600% on a target area of the HDV sample which we know has been through the down-conversion process due to the HDV codec. As I examined the pixels, both closely and at a distance, I saw vertical lines. I highlighted them to reveal that they were evenly spaced at every 4th horizontal pixel, which happens to correspond to pixel interpolation of every 4th pixel associated with a 25% horizontal pixel contraction and expansion in the conversion process. Then I examined the same area on Derek's uncompressed sample and did not see any evidence of vertical patterns, which is a very good thing because that means down-conversion before HDMI output is not evident within these samples. I uploaded the highlighted comparisons to http://thomann.net/hv20/interpolation/ EDIT: What is most notable about these comparisons is that each pixel within the uncompressed sample sharply appears to represent a single individual color on its own, especially when compared to HDV. That is some incredible footage. I am extremely impressed. EDIT: I have also examined NEOHDV against NEOHD and Uncompressed. As expected, since NEOHDV is 1440x1080, the same lines are evident every 4th horizontal pixel; whereas they are not evident with NEOHD and uncompressed which are full 1920x1080. This is further evidence that HDMI out has not been down-converted to 1440x1080. I have added these comparisons to the link above as well. Last edited by Mike Thomann; May 15th, 2007 at 11:06 PM.

May 16th, 2007, 12:19 AM	#24
Mike Thomann Regular Crew Join Date: May 2007 Location: Amherst, NY Posts: 100	When having to remove and re-create every 4th pixel, the only option is to take an average of the pixels around it. The only variation is in which pixels you take an average from. Photoshop provides a choice of nearest neighbor, linear, and bicubic. I have resized the uncompressed sample to 1440x1080 and back to 1920x1080 with Photoshop using those three resampling methods. I then did a layered comparison of those to the original uncompressed sample. Bicubic and linear were nearly identical to each other as expected in a strictly horizontal resize and both appear to impose quite a bit of blur on the image on the pixel level, and without comparing to their original source there are no signs of downrezzing. After looking at them, the uncompressed appears 3D like and razor sharp. The nearest neighbor resampling method on the other hand showed the same pronounced evidence of downrezzing as HDV and NEOHDV, but this method appeared just as sharp as the uncompressed but with less distinguishable detail. It may stand to reason that IF the HV20 was to downrez before HDMI output, it would do so with this method and hence we would see evidence of it. There is a chance that the HV20 uses a resizing method such as linear or bicubic which would leave little evidence of resizing. When looking at the amount of blur they generate on the pixel level I'm unable to see how Derek's uncompressed sample could possibly be the result of such a method. There's just too much blur taking place that spans all the pixels to the point that it appears as though you're looking at a compression codec. I'm not typically one to come to such conclusions but it's obvious to me that Derek's uncompressed sample has not been through such a process, and I'm happy with that. In the end, even if resizing was taking place, at these resolutions I need to view it a foot away with at least 300% zoom to begin to notice and in the real world I would need to be within 10 feet of a 55 foot screen. ...but like I said I see it as obvious that no resizing is even taking place. Last edited by Mike Thomann; May 16th, 2007 at 01:19 AM.

May 16th, 2007, 10:08 AM	#29
David Newman CTO, CineForm Inc. Join Date: Jul 2003 Location: Cardiff-by-the-Sea, California Posts: 8,095	>If it is obvious at 200%, then it might stick out in an theatre. He didn't say it was obvious, the compression is light only subtle noise variation is visible in magnified views. We aim, just like uncompressed, to store everthing, noise and all. Also theatre presentations are easiest to achieve a quality result, we passed all those test years ago. Just consider RED's snafu, the 4K presentation looked great, yet when people got to see the stills they saw compression issues, which is an issue for effects work, now the camera is delayed. It is much harder to do compression for the compositor -- this is the area we have grown. Read more here : http://cineform.blogspot.com/2007/04...eform-444.html Anyway, aren't we talking about the HV20 here? Are we really planning theatre presentations? :) > Has anybody tried the trial version of this lossless codec, averaging over 6:1 compression Lossless compression only averages 2:1 from any vendor, the noise component of film or video is too great for much more compression. Your idea (which we discussed before) of extracting the noise, and compressing only the "signal" is an idea that can't be reached to the industries satisfaction -- always some signal is lost with the noise reduction. This is why highend is so pro uncompressed, signal + noise, decide later what to do with it. Yet uncompressed is such a pain, and the industry has generally accepted HDCAM-SR level compression, this is the compression quality target for us: store everything such that you couldn't tell which was the source. More here : http://cineform.blogspot.com/2007/03...en-screen.html Now back the the HV20 + Intensity thread. __________________ David Newman -- web: www.gopro.com blog: cineform.blogspot.com -- twitter: twitter.com/David_Newman

May 15th, 2007, 03:47 PM	#19
Noah Yuan-Vogel Major Player Join Date: Dec 2004 Location: New York City Posts: 613	So if only the cineform ones were inverse telecine'd, theres a good chance the MJPEG one was still telecined and the screen grab ended up happening on a hybrid frame? maybe thats why there seem to be interlace artifacts. David's rescale of the neo hdv frame certainly does look much better than the previous version of that image. Any idea why it's so different?

May 16th, 2007, 12:11 AM	#23
Terence Krueger Regular Crew Join Date: May 2007 Location: toronto, canada Posts: 65	"This is further evidence that HDMI out has not been down-converted to 1440x1080. I have added these comparisons to the link above as well" it could also just mean that the camcorder is better at resampling the footage than whatever editor derek is using since the hdv and neohdv are actually 1440x1080 on disk. i think the footage looks great regardless. terence

May 16th, 2007, 05:41 AM	#25
Glenn Thomas Major Player Join Date: May 2005 Location: Sydney, Australia Posts: 320	David, I'm quite impressed by your Neo HD codec, but at $599 its more expensive than a copy of Vegas+DVD, and is also the same price as a MacMini. At that price I just think you could be excluding a large portion of low budget indie film makers who are just starting out. If they're working with the HV20 for instance, they're going to need a fast computer to start with which may cost as much as the camera. Then an Intensity card if they plan on capturing high quality (1920x1080P) footage using HDMI, which I've seen for $250. Finally the NEO HD codec which is more than double the price of the Intensity. Don't take this the wrong way as I realise it's a great product and I'm sure you'll sell a lot to more professional film makers, but a lot of people coming from say a DV background just aren't going to be able to fathom paying so much for what is essentially a capture codec. Perhaps you could do a cheaper cature only version without all the conversion features?

May 16th, 2007, 07:58 AM	#26
Thomas Smet Trustee Join Date: Mar 2004 Location: Milwaukee, WI Posts: 1,719	Glenn, High quality comes at a cost. That is why there is NeoHDV which is much cheaper. If you are really on a budget you could always use the free mjpeg codec that comes with the Intensity card. In my opinion the fact that you can now get a camera that can shoot 24p for $1,000.00 compared to in the past needing to buy a DVX100 for over $3,000.00 should mean many indies should have about $2,000.00 left over to buy a new computer and software. Everything is already much cheaper then it used to be. With SD 24p many of us had to spend a lot of money to get higher then DV quality. Even if you wanted to capture live uncompressed the only option we had was a Y/C S-video output from our cameras. The only camera I had that could pump out close to uncompressed SD was a $10,000 DSR-300 with a $80 26 pin connector that allowed component output. Now for $1,000.00 + $250.00 we can capture pure uncompressed or lightly compressed HD. There is also the fact that there really isn't all that much extra detail when you capture 1920x1080 from pretty much any HD camera. You will only notice a very tiny amount of sharpness between 1920 and 1440. If somebody really insists on having that tiny edge in detail then they should be ready to pay for that tiny boost.

May 16th, 2007, 08:51 AM	#28
Wayne Morellini Inner Circle Join Date: May 2003 Location: Australia Posts: 2,762	David, for your codecs and software, can I suggest an cleanup/film processing tool. Basically cleans artifacts, restores image, removes noise from compressed images like XDCAMHD HDV, MiniDV, AVCHD, h264 cameras (like the recent 720p $299 h264 pocket camera, and Sanyo in news). Both applied post, and automatically on capture. Back to HV20 Intensity.

May 16th, 2007, 11:06 AM	#30
Noah Yuan-Vogel Major Player Join Date: Dec 2004 Location: New York City Posts: 613	from the sound of it, that microcosm codec might be too slow for realtime capture, and isnt much more better at compressing than bitjazz sheervideo. Looks like for live video it only gets around 3:1 compression and only supports RGBA.