FX1 and simple FX in Cineframe24. My experiences

[Steven White]

About 3 or 4 weeks ago, I got my grubby little hands on an FX1. Since then, every weekend I've had the opportunity to shoot stuff, and have been having a blast.

Aside from what I read on this forum, everyone who I've shown Cineframe24 footage to has had absolutely no complaints, either live from the analog out, or as playback off of tape. I've shot a variety of fast moving objects, from nunchucks to waterfalls to my girlfriend tossing her hair in parody of so many makeup ads, and really, when played back on its own looks fine to me.

Next to Cineframe30 or 1080i footage it obviously does look choppy, but I'm not sure if that's the film effect or not. Obviously people here don't think so.

I shot a (Star Wars *sigh*) action short several weeks ago in Cineframe24 with a 1/60 shutter. I downloaded the footage from the camera twice - once to proxy DV files, and once to HDV. I then performed a preliminary edit in Premiere Pro 1.0 on the DV files... and was then ready for FX work on the HDV.

After reversing the pull-down to get a clean 23.976 fps in After Effects 6.0 and saving the 24p file to an intermediate HuffyUV file for faster renders, I begun rotoscoping the lightsaber effects 2 days ago.

In After Effects with regular zoom levels of 400x some artifacts ARE visible in the form of colour bleeding around edges, motion blocking, and (I think?) deinterlace errors from the Cineframe24 process. The potential deinterlacing ghosts occur by fast moving objects in front of bright backgrounds and are extremely subtle - and likely inperceptable if I wasn't examining the footage so closely.

The last paragraph however shouldn't raise nearly as many red flags as its going to. I have never worked with footage this crisp and this easy to track.

Following rotoscoping and preliminary compositing, the 16:9 image was cropped to a 2.35:1 file with a 1440x817 resolution (PA 1.33). For export to show my friends I rendered to a 24p 1440x613 Sorenson3 Quicktime file using 50% quality settings and keyframes every 48 frames.

Here are two clips:

http://s94963366.onlinehome.us/Eclipse/svjFX1.mov (2 MB)

http://s94963366.onlinehome.us/Eclipse/neckFX.mov (1.6 MB)

In the second clip, additional blur occurs in the form of a virtual pan (no zoom though). Compression rates are about 1 MB per second.

Comments and criticism are welcome - though keep in mind a lot of the compression artifacts that you see in THESE clips are a result of the Sorenson3 - not the preliminary MPEG-2. You may also consider playback at 1/2 size (or more accuractly 60%) to be an adequate representation of how this footage would look on a DVD - though I would never transcode at a rate lower than 9 Mbps (given how short the film is).

EDIT: Oh - and by the way - the time saved rotoscoping only 24 frames as opposed to 30 is greatly appreciated.

EDIT II - and please don't post links to this topic elsewhere. I have limited bandwidth, and if posted at certain Star Wars boards I could get disqualified from the competition this is for... which isn't a good thing.

-Steve

[Hayden Rivers]

Wow, you're really good at that rotoscoping.

None of that footage looked bad (although I might just be distracted by the lightsabers to notice the finer details). Someone else posted some CF24 footage at another thread which looked dreadful.

I really liked the neckfx.mov clip. It's interesting to me how much "space" the higher resolution and 16x9 allows you. It's kind of hard to explain but I think we'll see some interesting narrative stuff come out of HDV.

Keep us up to date and let us know when it's finished. I will pretty much watch anything that has a good looking lightsaber battle.

[Ignacio Rodriguez]

Could there be a difference in the quality of the the Cineframe firmware among initial cams? I suggest people doing Cineframe tests include the serial number of their cam and/or if possible the firmware version.

[Evan Estay]

good to see people are making some progress with cineframe. However, the clips are really short....i know you are involved in some kind of competition but if you could provide some longer clips I wiould appreciate it =]

as for mr.rodriguez, are you serious? this camera just came out....somehow I doubt that someone who bought the camera yesterday will have a better cineframe mode then the camera i bought 3 weeks ago.


Anyways, I look forward to messing around with cineframe24 (now if only i had some way to edit hdv footage)...i have feeling it will prove useful and tasteful in the right situation.

[John Gaspain]

man o man...your CF24 looks ALOT differnt than mine...mine looks more like 4 frames per second not 24.

Congrtatulations on your success with 24 FPS!!

I must know why though...

[Hayden Rivers]

Seriously, why does your CF24 look better than everyone else's?

[Mike Tiffee]

<<<-- Originally posted by Hayden Rivers : Seriously, why does your CF24 look better than everyone else's? -->>>

Because he removed the 3:2 pulldown which caused that stuttery look. The movie you viewed was at 24fps with no pull down.

[Alex Raskin]

Other thread explained that there should not be any 3:2 pulldown removal for Cineframe 24 or 30 footage, because Cineframe is not really 24fps but rather a simulation.

Am I reading it correctly that people who actually work with such footage professionally *do* find a need for 3:2 pulldown removal?

Why difference in opinions?

[Barry Green]

Because he removed the 3:2 pulldown which caused that stuttery look. The movie you viewed was at 24fps with no pull down.

Removing the pulldown does nothing to change CF24's stuttery look. Logically it shouldn't, and in practicality it doesn't.

CF24 looks stuttery because it is dropping frames/fields in its attempt to simulate a 24Hz capture rate. If there's not much motion difference between frames it can look okay, but if there's something moving quite fast, the difference between it and genuine 24P are glaringly obvious.

CF24 doesn't sample motion at 24Hz, it samples it at 60Hz and then drops fields and combines fields from alternating frames to deliver 8 out of 10 fields to the tape. The result is that you have uneven motion sampling, which is quite jerky. Removing the pulldown gives you 24 distinct frames that also exhibit the same jerky motion.

[Ignacio Rodriguez]

> CF24 doesn't sample motion at 24Hz, it samples it at 60Hz
> and then drops fields and combines fields

If so, it's a pity because if this cam actually captured at 24 fps, it could help cover one of it's major shortcomings, the low sensitivity inherent to it's small, high resolution (hence small pixel) sensor design.

[Steven White]

Hey Pixel - yeah, it's for the TFN Lightsaber Choreography Competition.

As for seriously why this Cineframe24 "looks better" than anyone else's, I can't honestly believe it does. But I do think without reference footage to compare to, and given the rather abrupt and erratic motion of the combatants in this footage, it would be very hard for the average brain to note the difference. You're not watching stuff moving at constant speed... so how can you rationalize a measurement of "stuttery"?

My philosphy on this kind of thing is that 9 out of 10 trained monkeys couldn't tell the difference... and I suspect off hand, that if I downsampled to DV resolution and shipped the same footage over to an XL2 board or a DVX board, they wouldn't even think to tell me it came from a stuttery FX1 unless I gave explicit reason to think otherwise.

-Steve

[Alex Raskin]

OK, so if CF24 is dropping frames, what does CF30 do?

[Dylan Pank]

<<<-- Originally posted by Alex Raskin : OK, so if CF24 is dropping frames, what does CF30 do? -->>>

Whereas CF24 drops the odd fields (or the even, I don't know which), then skips every 6th whole frame, then merges 4 frames into 10 fields applying a 2:3 pulldown, CF30 merely does the very first stage of dropping every other field. Every even (or odd) field scanned from the CCD is transferred to it's own frame.

[Alex Raskin]

Dylan, so then CF30 mode only results in 50% of the image resolution?

[Dylan Pank]

<<<-- Originally posted by Alex Raskin : Other thread explained that there should not be any 3:2 pulldown removal for Cineframe 24 or 30 footage, because Cineframe is not really 24fps but rather a simulation. -->>>

Alex, What there can't be is the kind of native or on-the-fly 2:3 removal that you can achieve with DV's 24p Advanced, because 24pA has a redundant third frame in the 2:3:3:2 cadence which contains fields from the proceeding and following frames. In that sense 24pA in the DVX100 or the XL2 is not really true 24p either, as it is still recorded at 60i onto the tape, though true on-tape-24p wouldn't offer any real improvement.

Some NLEs (Avid Xpress Pro, FCP and I think Vegas) can skip this frame during capture or even in playback and adjust the speed from 29.97fps to 23.976 (or even true 24fps). This can be done without re-compressing the footage because DV is all I-frame compression.

Now, even if the FX1/Z1 did offer a 2:3:3:2 pulldown (is does not) then it would not be able to do this 3rd frame removal because mpeg2 compression relies on all the frames in the Group Of Pictures (GOP).

However if you are reencoding the movie to another file, then the software can remove the 2:3 pulldown, though basically the footage is not really HDV anymore, though it may still be High Definition (as the case with Steve's footage: it's Hi Def quicktime, not HDV).

And yes, CF24, CF25 and CF30 all reduce the frame resolution by half.