AIC and HDV
 

is the apple intermediate codec GOP based or frame based? and is it inferior to HDV? It's just that even though I have a relatively new powerbook, it takes a long time to render GOP based footage. Just wondering if people out there use AIC?

It is frame based and inferior to native HDV. I doubt rendering AIC material will be faster than HDV except for Print to Video.

I disagree with Ben. Version 1.0 of AIC may have had some problems, but version 1.0.1 is in no way inferior.

AIC is considered a "lossless" codec. It is not uncompressed, but the idea is that no new artifacts will be introduced, and the original quality of mpeg2 will be maintained.

Here's my proof:

Right-click here to download a TIFF exported via Mpegstreamclip from the original captured m2t file.

and then right-click here to download and compare this TIFF exported from Quicktime Player Pro of the AIC converted version (via Lumiere HD) of the same shot.

I chose this frame because it gave the Mpeg encoder a workout with the bright flashes and the dark areas, and smoke causing banding with the 8-bit bit depth. Therefore I consider this a "worst case scenario."

Also, AIC will render effects way faster.

Hey Tim nice shot of the soldiers, but unfortunately the second file can't be opened it says it's corrupted, the AIC shot.

BTW I'm going to use your reversal/filmvert this afternoon on a shoot, I'll let you know how it goes.

Tim is it easier to render on AIC than HDV?

Thanks

Hmm. I just downloaded it and opened it with preview without a problem. Are you trying to view on a PC? I'll open the file in photoshop and resave it as a tiff, maybe something isn't flagged right from the Quicktime export. So try again, then open both and compare. They are virtually identical except that the gamma in AIC seems to be set correctly.

Cool. Start a new thread in the HD100 forum.

In my experience, on my dual 1.2Ghz G4, AIC rendering is WAY faster than native HDV.

Tim,

Your screen grabs do a great job of show how well the AIC codec works. With tabbed browsing in Firefox, the files open with the QT browser plugin and you switch between tabs to see some very slight differences. I actually like the AIC shot better. It seems to be a slight bit darker which is how I tend to color correct any way.

I dropped the AIC frame on top of the original in difference mode and the result is here. While the 2nd image indicates that there are significant differences between the two, the first shows that they are for the most part visually imperceptable. The only place where they might be visible is in the coronas of the muzzle flashes - seems to be significant difference in the blue channel only.

I'd say it's an insignificant difference considering the performance trade-off, especially on a machine with a slower processor.

I'm pleased that AIC works for you, but you can't compare both based on a single frame. The way I understand AIC is that is an MPEG 2 like I-frame only codec. This means that HDV I-frames get transferred to AIC without recompression. It is the B- and P-frames where you will see additional compression, and where the losses are. I'm not sure but then again, I think that nobody, besides Apple, knows what AIC actually does.

I admit that if you're going to use an intermediate codec AIC makes more sense than DVCProHD. The thing I hate about AIC is the change in workflow when capturing.

I've never seen losses or additional artifacts on any frame, and I checked many different clips side by side, frame by frame, before I committed to to the workflow for one of my current projects.

I've put together a difference-mode test here with a video clip:

http://www.divergentshadows.com/hdv/...nce_video.html

The original test video appeared pure black, it took a fairly extreme application of a level filter to make the difference as visible as it is in the clip above. You can see that the I frames exhibit the least amount of change between the two, although I'd estimate that over half of the pixels in even the I frames are different. Either way, the differences are so minute that they are unlikely to cause any visible difference unless you plan on running through multiple generations of the conversion.

The video went from 3.2 to 10.3 MB/second in the conversion to AIC - apparently this can vary though depending on detail, motion, etc in the original clip, and tops out at about 12MB/second for 1080 HDV.

Very interesting. Is it correct to say that your video displays differences from 0 to 5 on a 0-255 scale? That is, white is a difference of 5 between AIC and HDV for a given pixel? Blue would be 1 "unit" difference, basically a bit error caused by rounding?

The "black" frames appear to be I-frames because they're strictly 15 frames apart.

There's one thing in your comparison which may be unfair to AIC: I'm pretty sure Quicktime player decompresses each HDV-frame to uncompressed before encoding it to AIC, in contrast to Final Cut Pro's AIC capture which has access to the raw HDV material. Can you try this again, but this time capture the same clip twice, once in HDV and once in AIC?

Actually it's probably closer to 0-11 - I just pulled it back up to check the exact values. Setting the black input and white input one level apart and progressing up the scale(i.e. 0/1, 1/2, 2/3, etc) shows all differences disappearing at 12/13, with probably 95% of the 3 channel differences (maybe 75% of the differences total) dropping off by 6/7 leaving mostly individual errors in blue & red with very few in the green channel. So yes, considering a 256 value scale we are looking at most potentially-visible errors falling into the range of 1-3% difference from the original source - again, likely below any reasonable threshold of perception unless you plan to run many generations of the conversion.

I hadn't thought of that, I was assuming that FCP would decompress the footage before the conversion - but if it is truly just transcoding the video stream it may drop the difference even lower. I'll run the test again with two different captures into the two formats.

Ok I just ran the test again, this time capturing the same clip as HDV and AIC in order to (potentially) improve the AIC error rate by transcoding rather than re-compressing...and there is good news and bad news.

First, the good.

Capturing HDV directly to AIC appears to be slightly better than re-encoding via QT pro. Using the same methodology as above I would say most of the 3-channel difference dropped off at 4/5 while the final remaining blue/red differences were gone by 9/10 - so this drops the (potentially) visible errors down to a max 1.5% difference from the original. So, technically it's a slightly better conversion, although neither one should result in any visible difference from the original.

Unfortunately, there's still the bad news - perhaps it's only news to me.

This was the first time I'd tried capturing to AIC, and I was surprised to find out it does not allow for device control - therefore you cannot log & capture nor can you batch capture. It works only in a "capture now" sort of mode where it starts playing the tape and captures continuously until you hit escape. It also fails to retain the timecode from the original tape, so you cannot batch capture later. To me this would essentially make it completely unacceptable for any serious project. Without the ability to batch capture later you cannot back up a project without backing up all the media, and you can't archive a project without archiving all of the media as well. Considering AIC takes 3-4x the disc space, backing up all of your source media is considerably more difficult. Am I missing something here???? Or is this really the way it goes?

I have a different workflow using AIC. I capture and edit in Native HDV using FCP 5.04. Then I render out a MASTER clip in AIC. As I understand it, FCP goes to 4:4:4 before rendering out. Using AIC, I figure this is a virtually lossless version of the HDV originals. If you render out to HDV it adds UGLY artifacts. I could go out to Uncompressed but it takes HUGE storage. That then goes out through my KONA LH card to whatever deck I need. If it's SD, the KONA will downconvert on the fly. I'm working on a big doc right now and will rent an HD deck for the deliverable MASTER tape version.

I see no need to go to AIC first, instead of editing native HDV. Even if you add Color correction, it goes 4:4:4 before the final render. There is a link on Apples webpage about editing native HDV and mentions this, it's a PDF file that has been referenced here before.

I've ran some tests and checked the results using "difference" in AE and this is the best quality/lowest storage solution I've found. As mentioned, this version of AIC is WAY better than the first one!

Paul,
I think you're right, AIC is probably the best mastering format (vs either HDV or uncompressed, etc). I think any conversion/render in FCP is done in a 4:4:4 space so it's definitely not losing much quality - it should match my first test which involved conversion from HDV to AIC and still produced no visible artifacts.

However, I think there are definitely arguments to be made for capturing to AIC - you get smoother playback/scrubbing/shuttling & more realtime effects/higher quality realtime previews vs. native HDV. For some people the tradeoff in workflow speed vs. disc space is probably worth it, especially if disc space isn't an issue or if working on a slower machine (a powerbook for instance). However, the timecode and capture issues pretty much make it a useless option for any serious project, at least until cheap blu-ray burners become available for archiving!

@ Paul Frederick:

I think your workflow is one of the better ways to archive HDV originated material. However, I'm curious if the quality of your AIC file played through the Kona is better than playing a straight HDV timeline with renders? I understand that the exported HDV file will be a lot worse than the AIC file, but playing directly from the timeline should be better 'cause you will drop one recompression step.

You might want to think about PhotoJPEG@75%. Graeme Nattress is a big proponent of it. I think the filesizes will be comparable to AIC and you have a 4:2:2 colour space (better for graphics).

@ Evan Donn:

That's what I ment when I said "I hate the change in workflow with AIC". I do not think that the difference between HDV and AIC is insignificant but I will have to do my own testing for that. Thanks for the valuable comparison and the time you put into it.

Ben,

For the work I'm doing I color correct every shot. I also often use dissolves. When I render in the timeline it renders to HDV (even though it does so in 4:4:4 color space) since that is what the project is set to for HDV native editing. I think it looks worse playing straight out through the KONA after this HDV render then if I use RT Playback while editing and make the AIC master.

This is my subjective opinion, its mostly visible in the blockiness of the dissolves.

I usually lock in the edit first, then add color correction, then render the project out to AIC. I will have to try the photoJPEG at 75% for a render and check that out.

Codec Test
 

I was recently trying to decide which codec was best for editing and color effects myself as I do a lot of CC in my videos. I did this simple test to find out which codecs could handle effects best from HDV source material. The test is here:

http://video.terpstar.com/extra/

I appologize for not having something exciting like soldiers to test on, but a white wall tends to show banding when using lesser codecs.

Clearly to me HDV is right out and AIC is still too compressed. DVCPro seems to be the right choice in spite of the smaller resolution. I would consider PhotoJPEG75%, but it does not want to play back at realtime on my machine (1080i) and therefore is not editable. I'm interested to hear what people think about this.

-Chad

Chad,

Very interesting results. There is a clear dilema going to DVCPro however as you noted, the resolution is lower. Now I think the test needs to be done with a resolution chart to see how it holds up at the lower esolution of DVCpro. Wish I had one on hand.

Paul,

I found a decent res chart online and proceded to simply export it from different timelines in Final Cut (AIC, DVCProHD, & JPG75). The TIFF files are available here:

http://video.terpstar.com/ResChartTest/

The AIC and JPGEG75 are pretty much indistinguishable, but of course when I stretched the DVCProHD image to 1920, it showed a little bit of loss. However this is a small trade-off to me when you consider the compression artifacts which AIC produces (as seen in my previous test). However, not wanting to really lose any resolution, I would be interested in an intermediate for the Mac which is similar to Cineform products. If anyone has heard of such a thing, be sure to post it here. Otherwise, I can't see any alternative to DVCProHD when working on a Mac since PhotoJPEG75% is uneditable at 1080p24 (at least on my 1.25Ghz FW800 machine).

- Chad

It seems to me what is going on here is an indistinguishable result using AIC with the JVC HD100 and terrible results using AIC with Sony Z1 or FX1. All of my personal tests have been with HD100 24P or 30P footage.

The question is: Is the difference in 6-frame GOP vs 15-frame GOP, or Progressive vs. Interlace, or both?

It is Interlaced vs. Progressive Tim.

AIC loves progressive footage. It hates Interlaced. The progressive footage from a HD100 and a XL-H1 looks indistinguishable when converted to AIC. But interlaced footage converted to AIC looks a mess.