HDSDI and HDV frame grabs

[Jason Varner]

You know what I'd really like to see is a TRUE HDSDI frame grab. Maybe it's just me but it's amazing that you got those chickens to freeze for long enough to shoot tape and then capture using hdsdi and cineform. You are truly a master of poultry. I could never get my chickens to stay still for more than a minute or so. Oh wait I get it, these are all hdv playback to different codecs. Well Alister my good man if you have the time and tools what I'm DYING to see is a true hdsdi grab. It's pointless to record hdv and then export hdsdi. Please be a pioneer and do what so far no other has done and bust a real time hdsdi mother lovin' 1.42Gbps capture, then we can talk turkey. Good Luck, Jason.

[Rabi Syid]

Originally Posted by Jason Varner

You know what I'd really like to see is a TRUE HDSDI frame grab. Maybe it's just me but it's amazing that you got those chickens to freeze for long enough to shoot tape and then capture using hdsdi and cineform. You are truly a master of poultry. I could never get my chickens to stay still for more than a minute or so. Oh wait I get it, these are all hdv playback to different codecs. Well Alister my good man if you have the time and tools what I'm DYING to see is a true hdsdi grab. It's pointless to record hdv and then export hdsdi. Please be a pioneer and do what so far no other has done and bust a real time hdsdi mother lovin' 1.42Gbps capture, then we can talk turkey. Good Luck, Jason.

And in a nicer fashion that would put the icing on my cake aswell!

[Shannon Rawls]

Originally Posted by David Newman

We have customers posting feature films and completely bypassing the off-line stage. <snip> People can now store 40+ hours of HD source material on disk space that will cost only around $1000, no elborate RAID required.

This is what I love most about CineForm for PC & DVCPRO-HD for MAC. No off-line proxy files are needed to edit then go back to online. Simply cut the HD footage itself (with no loss of quality after you apply effects, transitions, etc...) and you're done. Same workflow as DV.

- ShannonRawls.com

[Alister Chapman]

Originally Posted by Alister Chapman

The SDI capture was direct, "Live" from the camera to the PC.

As I stated earlier the HDSDI feed was direct from the camera to my PC via a decklink HD SDI card. There was no compression used. Getting three identical frames was easy, I recorded to HDV tape in camera at the same time as I fed the HDSDI to the PC.

I then transfered the HDV material to the PC using connectHD which I had set to both capture the native m2t file as well as create a cineform avi file.

I then opened a 1920x1080 uncompressed project in premiere imported the three clips, found the same point in time on each and exported the stills. I deliberatly chose a frame with little movement as interlace artifacts on the stills would detract from the subjective image quality.

Simple really.

[Vince Gaffney]

I'm must be missing something here. I have the Decklink HD card. The Basic one. I demo-ed the H1 here for several hours looking mostly at workflow options (the flexibilty is amazing) We could not get the 60i HD SDI signal into FCP. And I tried about 20 set-ups. Did I miss a step somewhere? Do I need to capture outside of FCP and import it? This may not be the right place for this, but I am truly confounded.

Vince

[Wayne Morellini]

Alister,

5MB per image is alright, nothing compared to the size of test clip's I've had to download.

Could I ask for a cineform compressed version of that HDSDi frame, it will be interesting to see how close the cineform frame comes to the original.

Separate frame files would be good for me, I don't have photoshop, I can flip between them full screen in Opera.

Thanks

Wayne.

[Alister Chapman]

Vince.. I am using a basic Decklink HD card, but on a PC and it works for me.

I have posted some more clips, a bit more colour in these, they are uncompressed Targa files that I have zipped to make them a little smaller. Again there is a direct HDSDI capture, the same frame off the HDV tape plus the cineform version of the same frame. I have also added a frame from my Z1 taken at roughly the same time.

Wayne.. there is also a 8bit Cineform conversion from the HDSDI stream as requested. All the grabs can be found on the link below.

http://www.ingenioustv.co.uk/tests/xlh1tests.htm

[David Newman]

Pity they weren't all of the same frame, but your samples clearly show the enhance HD-SDI sharpness (whether CineForm compressed or not.)

[Vince Gaffney]

Alister,

Thank you. I had a camera in for a few hours yesterday and ran out of time to really work with the direct capture of the SDI feed. Mine should be here Monday or Tuesday and I will try it all again. I got a few tips from Decklink Support that may help as well.

Vince

[Wayne Morellini]

Thanks Alister. I am downloading now. I mainly want to flip back and forwards between the cineform and the HDSDi to see how visually lossless it looks. I've viewed the HDV at a FOV approximately Equal to a good cinema seat 2/3rds the way to the front, and it looks very smoothed out and videoish at that size.

David, this visually lossless, what does that mean, is it somewhat lossy, or is it true lossless but compressed over sequence of frames? Is it a derivative of h264?

There is a shortage of cheap component/HDSDi recording solutions at the moment, have you guys considered getting your codec on compression chip that could be used for such a solution? Have a look here (It's Keith's project):

http://www.engr.mun.ca/~wakeham/

Thanks Again.

[Wayne Morellini]

I've done the cinema testing, not to easy as Quicktime was the only thing that would handle TGAs on my system.

At the Field of View of a good cinema seat, the HDV frame clearly looked out of focus compared to the uncompressed HDSDI frame (caused by HDV not focus) and actually looked a bit videoish and detail smoothed out in comparison. Which is a real surprise, as the scene has very little movement to trip up the codec and lower quality, so the two frames should look a lot closer together. It is possible that professional post processing in a film transfer lab could sharpen up the edges, but it is still smoothed out detail wise. Colour looked neutral on the HDSDi, and out of whack on the HDV frame. Good enough for TV work and low FOV (not getting into the motion artifact issues).

In comparison of the HDSDi frame to the cineform compressed HDSDi frame, the out of focus effect was still there but much less. Looking at the bricks under magnification, or the grain on the wooden fence and nails, it was easy to match up detail. There were differences in the detail, but the detail was still there. At Cinema FOV, the detail and out of focus effect from the compression was barely noticeable. Both were much more pleasant and relaxing to look at than the HDV frame.

I have seen better frames from the higher quality Bayer Raw cameras we worked on in Alternative Imaging (one was a dog) such breath taking life like detail. What are the HDSDI frames formated in this case, 4:2:0, 4:2:2, or 4:2:0 footage pout over a 4:2:2 format? Whatever it is, it doesn't look 4:4:4, and the 4:2:2/4:2:0 format clearly is detracting from the quality to be gotten from 4:4:4. Maybe it is the actual sensor resolution pixel shifted and up-converted to 1920*1080 res that is doing it. What ever it is, it doesn't have the surreal presence, and depth of detail we got from Bayer Raw.

Now the big question. If HDV can't get near the quality of a low change/movement uncompressed frame from a locked off shoot, which probably indicates it is loosing a lot of resolution compared to filming a simple (to compress) resolution chart, how low is the resolution getting during frames with movement or large image changes?

HDV has been caught out. If you want stunning looking images for the big screen, then uncompressed/lossless or uncompressed to Cineform compression is the way to go. If you want HDV, get stunning content.

Now the other question, how does cineform compressed HDSDi compare in frames of high movement and change, like surfing tight shots. Any chance of getting those chickens to dance Alister ;) .

Alister, if you are interested in HDSDi for your filming you might be interested in this, a cheap convenient way to capture:

http://www.engr.mun.ca/~wakeham/hdu1.html


Thanks

Wayne.

[Alister Chapman]

Interesting observations Wayne.

What I would just point out is that the Cineform codec I have is the standard 8 bit codec which is limited to 1440x1080 resolution so I am not surprised to see some softening. I would love to have the full 10bit Prospect codec at 1920x1080 as I belive that would be the way to go. Having said that all the dlivery formats I intend to use are 1440x1080 so I don't see it being too big an issue.

As for overall picture quality, well I have spent some time with the H1 and a scope and SDI monitor and have found that the settings I used when I shot the clips was far from perfect, in particular the edge enhancement. It has been sugested by some that the H1 has to much edge enhancement at the default setting, so based on that I had backed off the sharpness some way. Having now looked at the output on a scope and decent monitor I would say that the "0" sharpness setting is pretty much spot on in so much as there is little to no overshoot on enhanced edges and to my eye a very natural look, less than 0 leads to rounded edges and a soft looking image.

The H1 lens isn't doing the camera any favors, maybe if this lens was on a Z1 you wouldn't be able to see some of the CA issues, but on the H1 you can see them. By the way the above shots were at f5.6 with ND2 in.

I belive that the H1 produces some stunning images, it is a shame that HDV is softening them so much. Perhaps later in the week I will get a chance to do a side by side test with an HDCAM, that would be interesting.

[Pete Bauer]

Wayne,

HD-SDI looked better than HDV? Of course. I guess I'm surprised that you're surprised -- although using Photoshop at 200% on dual 1600x1200 displays to view the grabs, I personally wouldn't describe the difference as dramatically as you portray.

I doubt anyone here thought that HDV (or capture to DVCProHD, for that matter) would match uncompressed (or Cineform) HD-SDI files. I'm just a Joe-Bag-o-Donuts shooter, and I sure expected it:

http://www.dvinfo.net/conf/showpost....87&postcount=5 (last para in particular)

No question again that large amounts of inter-frame motion differences can "break" HDV, but taking GOP beyond the brink (which, as it turns out, isn't really a practical concern for MOST shots anyway) has nothing whatever to do with a slightly reduced amount of detail, or variances in color space rendering, as compared to a "big brother" format like HD-SDI. I don't see the logic there...I think it is a red herring.

FWIW, I tend to get skeptical when folks are trying to convince me of something by using entirely vague and subjective terms like "video-like" or "film-like" that get thrown around all the time.

I don't think this very limited look means "HDV has been caught out." It only means what we all expected: compressed 1440x1080 at 4:2:0 isn't quite as sweet as lossless 1920x1080 at 4:2:2. At the end of the day, folks tend to set aside hyperbole and use what works best for their situation.

[Wayne Morellini]

Alister, I see what you mean, that must be why the HDSDi was so far off of the RAW images I have seen. But for my other observations I used the HDSDi footage as the purest one to compare them to (are you saying that the camera does true 1920 pixel resolution, and not 1440 resolution up converted to 1920). Unless the sharpness control effects HDV more than HDSDI, I doubt it makes enough of a difference.


Re-edit:

This gets technical, so you may want to skip it:

The zip files you sent are compressed lossless, so the ratio of the zip to the TGA file size may give a bit of a approximation to how much detail is in them. But this is not accurate, as the TGA files are padded with extra stuff, not all things compress equally, and some of the data may represent inaccurate image renderings from the codec itself.

(All figures approximate, results are approximate as well because of internal Zips structures)

The TGA files for the Canon are 6076K
The Z1 files are 4557K

houseHDSDI
6076/4207K=1.444259

HDSDICineform
6076/4200K=1.446666

housem2t
6076/3724K=1.631578

Housecineform
6076/3428K=1.772462

sonyz1
4557/2993K=1.522552

The differences in picture information is probably much more than what these figures indicate, because even if the information is wrong in an area there is still pixel data there that would minimise the difference. So the HDSDi and Cineform are very close together, I doubt it is as close as the figures indicate, because of the way zip compression works). There is significant loss in the HDV file, and more again in cineformed HDV picture (I didn't mention that the cineform of the HDV looks softer to me). But the interesting thing that caused me to post this is the Z1 result compared to the canon HDV . I know the two pictures can't be the same, and the different shadow content/sharpness in that picture could explain it, but is the Sony codec getting more image detail, it could just be those zip structures inflating the size again.

To accurately measure something like this, you would need to record uncompressed HDSDi, play back and record on both cameras through component (do they have component in?). Then pull the same frame from both, and compare it with some sort of metric to the original, like the SNR metric, or use a lossless video compressor (or look through visually). Would be interesting, might be splitting hairs, it's 1/16th difference, maybe the difference gets more, but the difference between HDSDi and m2t is around 1/8th, so the difference could be much more than 1/16th in reality. With a HDSDI camera a reviewer would be able to test the codec performance on any scene against the original, exciting.

Thanks

Wayne.

[Wayne Morellini]

Thanks Peter, but this has to be read in context, it is getting past the "hyperbole" of how "good" HDV is and using it for film. I'm not saying that you shouldn't HDV a movie, just be aware of the compromises and advantages you can get going elsewhere.

I used the still situation, because it represents one of the few situations where the codec can reach it's almost full detail, but was far from it, that was the genuine surprise, it should have been better. Using some calculation you can see what might happen to detail and resolution (because dropping back to bigger pixels is one of the prime ways of the codec dealing with to much information) given an action scene. So most scenes might (will) suffer this, not getting the frames full potential, in varying degrees worse than this. And, sure enough, how often do we hear, why does everything on my ... (Mpeg2) camera go so blurry when it/things moves (and we are not talking about normal image blur).

What we are left with, if viewed from a good cinema seat) is something that will look out of focus and smoothed like video (if you know the technicals of why video used to look like that, you know what I mean). Not being overly dramatic but realistic, given better eyesight than mine, which I'm sure that a large number of people in the audience have, it will be obvious enough to them. At the correct FOV, it starts looking like a CRT SD TV picture playing minidv (in detail and smoothing rather than colour, which is better) because the pixels look many times bigger than at TV viewing distance. But it depends on how you set up your viewer. Many people might let the software viewer shrink the image to fit the screen, or smaller again, this will result in pixels being combined together, hiding the problems. If they view it at a normal computer viewing distance the pixels may look 9 times smaller than on a cinema screen from a good cinema seat, hiding the problem. A monitor set too bright (blooming in the eye and on screen) or dark, or some LCDs, will also hide the problem, even how sleepy/well a person is.

I don't know the method you use for FOV viewing but this is what I do. Take a pair of glasses into a cinema each time and measure the amount of the lens the picture occupies from best seat position and back (look at the screen square on with one eye open fixed on the middle, using peripheral vision to measure lens frame intersection). These FOV are pretty standard, because your eye has zones of vision, when the screen is so wide that it crosses into purely peripheral vision you find yourself looking at the sides of the screen to catch things you are missing (unpleasant). For a medium framed glasses you will find that the seats range half a frame to almost a full frame on a large 6cm glasses lens. Now do the same at home, moving closer to your monitor till you get the video picture to occupy the same lens coverage (equivalent FOV) as the cinemas screen at the best seat (when this looks good all the rear seats will also look good). If it goes off the screen calculate and include the extra distance. This is the sorts of lengths you have to go through to get a good idea of what your audience will see, how they will see.

re-edit: With what Alister just said, maybe the blur is low enough in HDV (but not some of the other things) but will mean that HDSDi/Cineform will be even better.

Thanks