HDV + cineframe 24 + Cinema tools 3 = perfect 24p?

[Jack Zhang]

If anyone can get perfect 24p out of Cineframe 24 through Cinema tools 3, post it here as H.264. I want to see it! If no one can, post the reason why you can't.

[Dylan Pank]

You can't get "perfect" 24p out of cineframe 24 because the FX1/Z1 CCD always scans at 60hz, and always interlaced, and never at 24hz.

Now with some filtering and work you could get acceptable 24p from an FX1/z1 but it would entail shooting in 60i and then applying a de-interlacer (such as DVfilm maker) rather than starting with CF24.

[John Jackman]

If you want actual 24P forget Cineframe mode. It is a shoddy simulation and adds nothing to the process.

You can get "perfect" 25p out of the Z1U by shooting in 50i and using a smart deinterlacer.

[Graeme Nattress]

CF24 drops vertical rez by a half, and drops every 5th frame. There's nothing you can do in post to get it looking decent again. On talking heads you might not notice, but on any kind of pan or action, you'll see stutter, and what looks like too high a shuter speed.

If you're an FCP user and want decent 24p from HDV, use my Standards Conversion or Film Effects. I just did some HDV conversions from 1080i to 720p24 and they looked great, which is more than can be said for CF24, pulldown removed or otherwise....

Graeme

[Peter Moore]

1080/60i to 720/24p conversion is imminently doable and looks great if it's done right. Not perfect, but close enough that no one would know it wasn't shot on 24p natively unless you were really horrible with the camer (really fast jerky pans will look bad, though fast smooth pans will still look pretty good.

[Alex Raskin]

Graeme, what about CF30?

I hear that in-cam deinterlacer of FX1 only loses resolution in the moving areas - is that true?

Also CF30 won't drop any frames, correct?

[Graeme Nattress]

From what I've seen CF30 just throws away a field without any intelligence in the matter. CF30 doesn't drop frames, but it severely limits post options. The best modes are 50i and 60i and that's what I'd stick to.

Graeme

[David Newman]

I think Graeme is being a little hard on the Cineframe modes. The assumption that an external smart deinterlacer will always produce a higher vertical resolution than achieved using Cineframe 24,25 or 30. This is not always true, an external deinterlacer has to interpolate information this is not present in the image, this can result unwanted artifacts, or simply a resolution that is lower than Cineframe. That said in most cases external deinterlacer will on average produce a higher resolution image. However, if you go with a Cineframe mode your vertical resolution is guaranteed with little to no post-processing needed, this is higher than just a single field from a typical interlaced stream, but not a high as potentially achieved through de-interlacing. Post de-interlacing will make your post-production more awkward as you should edit in a progressive mode, meaning all you source materials must be de-interlaced first. Editing in interlaced then de-interlacing later doesn't lead to as good results -- transition and titling elements are compromised.

Using Cineframe you will get a vertical resolution of around 540 lines (as measured by test pattern charts.) This will be independent of the source image. Each frame will contain motion blur that matches your camera's shutter setting. You will have an easy post-production work-flow, enabling progressive editing immediately (if your NLE allows.)

Using a de-interlacing approach you will get a vertical resolution between 400 and 770 lines, depending on the detail in the image and the types of motion present. Deinterlacing can impact motion blur as two time slots are being used to construct one. For the highest quality you should deinterlace your source data (potentially that means a lot of footage from a high shoot ratio), each scene may require tweaking the de-interlacer's settings (plus each de-interlacer has different results -- making a lot of work for the best output.) Then you can start progressive editing.

If I was shooting for film, with a real budget and the time, I do favor the de-interlacer approach. However for the bulk the work, for any no/low budget work I would use the Cineframe modes. It is simply too hard for your the audience to tell the differences (particular if the story is compelling.)

[Alex Raskin]

Originally Posted by David Newman

Using Cineframe you will get a vertical resolution of around 540 lines (as measured by test pattern charts.) This will be independent of the source image.

OK, the way I understand it is this:

Cineframe 30 does NOT do smart deinterlace, but rather it DOES drop resolution in half for the entire frame, not only for the moving parts of the image.

Correct?

[Steven White]

You are correct. However:

There is more data to compress in a 60i stream (more resolution from CCD in combination with higher bandwidth information). Therefore, Cineframe modes are easier on compressors.

A single field in a 60i stream has less than 540 lines of actual resolution, because of vertical low-pass filtering to remove artifacts in interlaced display. Hence, straight (non-adaptive) deinterlacing of 60i leads to less vertical resolution than CF modes will achieve.

David's post is exactly how I feel about the matter. Shooting with CF modes is painless, and the resulting workflow is much much much easier than dealing with interlaced images. Render times are significantly reduced, and the whole post-production process is identical to what you would expect from true 30p and 24p source.

-Steve

[David Newman]

Alex,

Yes and No. For interlaced image acqusition the fields are typically filtered to reduce interlaced flicker, so each field will have only a vertical resolution of 400 lines. Yet the Cineframe modes have a higher resolution than this. The camera in Cineframe mode does no (or little) filtering from the interlaced field data and interpolates the second field from this image. The result is a measurable 540 line vertical image, which is not too bad. So this is not "smart" in the post-deinterlacing software sense, yet it results in a reliably good image. Any smart deinterlacing can result in artifacts with a variable output resolution (dependent on content.)

[David Newman]

Steve,
I didn't see you post before my reply. We are saying the same thing. :)

[Alex Raskin]

Thank you all for the very detailed explanation!

I'm primarily shooting in CF30 mode, based on the asumption that it doesn't lose much resolution while providing easy workflow with progressive images.

Just wanted to double-check that assumption, and seems like it holds.

Except I thought that CF30 did smart (motion area only) deinterlacing like the software-based solutions, and from your answers I see that it's not so - but still OK.

Here's another one for you. Which one is better for *bluescreen* work - CF30 or 60i? I do 60i as previous consensus seems to suggest, but what do you guys think?

[Steven White]

Blue screen work requires accurate colour information. The per-pixel accuracy of the key will depend on the per-pixel quality of the colour information, as it pertains to the data in the image. Ideally each pixel has data for all three primary colours (i.e. 4:4:4 sampling).

HDV decimates colour information with the 4:2:0 sampling and the interlaced compression. 4:2:0 means each line of the image has different chroma information, but that 1/2 of the chroma information for a pixel comes from the lines above and below it via interpolation. This is particularly troublesome when you consider an interlaced image, as the lines above and below any given pixel are acquired at a different time (this is probably why WMV9 uses 4:1:1 for interlaced and 4:2:0 for progressive).

I do 60i as previous consensus seems to suggest

I've never believed in this argument or the consensus from a mathematical standpoint (I do NOT have personal experience keying with keying - YET). Here's why:

The FX1 is a 3 CCD (green-1/2-pixel-shifted) camera. The maximum colour sampling it is capable of is 1920x1080 4:2:2. This effectively gives you a 960x1080 grid with full 4:4:4 colour to work with. If you take this 960x1080 grid and compress it to 1440x1080 4:2:0, you end up with the aforementioned colour problems associated with 4:2:0 and interlaced display. If you take the 960x1080, and toss out one field to make it 960x540 4:4:4 and compress it to 1440x1080 4:2:0, the colour information for the effective 960x540 grid will be MORE accurate than the 60i case.

In summary, I would expect that Cineframe30 offers softer images that are more easily compressed, with more accurate colour information per-pixel at the effective resolution of the image. Alternatively, I would expect to get a better key from a Cineframe30 frame than from a 60i field.

-Steve

[Shannon Rawls]

CF30 is very nice. I shoot this mode for action narratives and music videos.
CF24 is nice in the hands of an editor who knows what to do with it. Otherwise, it's a mess. I shoot this mode for my dramatic narratives
60i....weddings and family picnics. *smile* However if you have some good conversion software like nattress or dvfilm or vegas6....you're in good hands and can do ALLOT of post stuff.

- Shannon W. Rawls